CTNF 18/519,326 CTNF 94198 DETAILED ACTION Notice of Pre-AIA or AIA Status 07-03-aia AIA 15-10-aia The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA. This office action is in responsive to communication(s): original application filed on 11/27/2023. Claims 1-20 are pending. Claims 1, 11, and 16 are independent. Specification The use of the term " Bluetooth " and " Wi-Fi " in ¶¶ [0132]-[0133] , which is a trade name or a mark used in commerce, has been noted in this application. The term should be accompanied by the generic terminology; furthermore the term should be capitalized wherever it appears or, where appropriate, include a proper symbol indicating use in commerce such as ™, SM , or ® following the term. Although the use of trade names and marks used in commerce (i.e., trademarks, service marks, certification marks, and collective marks) are permissible in patent applications, the proprietary nature of the marks should be respected and every effort made to prevent their use in any manner which might adversely affect their validity as commercial marks. Claim Objections 07-29-01 AIA Claim s 2-8 are objected to because of the following informalities: in Claims 2-8, lines 2-3 , " … further includes instructions configured to cause the processor alone or in combination with other processors to perform operations of … " appears to be " … includes the excusable instructions configured to cause the processor alone or in combination with the other processors to perform further operations of … " to distinguish with " instructions " to " the generative model " recited in Claim 1 . Appropriate correction is required. 07-30-03-h AIA Claim Interpretation 07-30-03 AIA The following is a quotation of 35 U.S.C. 112(f): (f) Element in Claim for a Combination. – An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof. The following is a quotation of pre-AIA 35 U.S.C. 112, sixth paragraph: An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof. 07-30-05 The claims in this application are given their broadest reasonable interpretation using the plain meaning of the claim language in light of the specification as it would be understood by one of ordinary skill in the art. The broadest reasonable interpretation of a claim element (also commonly referred to as a claim limitation) is limited by the description in the specification when 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is invoked. As explained in MPEP § 2181, subsection I, claim limitations that meet the following three-prong test will be interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph: (A) the claim limitation uses the term “means” or “step” or a term used as a substitute for “means” that is a generic placeholder (also called a nonce term or a non-structural term having no specific structural meaning) for performing the claimed function; (B) the term “means” or “step” or the generic placeholder is modified by functional language, typically, but not always linked by the transition word “for” (e.g., “means for”) or another linking word or phrase, such as “configured to” or “so that”; and (C) the term “means” or “step” or the generic placeholder is not modified by sufficient structure, material, or acts for performing the claimed function. Use of the word “means” (or “step”) in a claim with functional language creates a rebuttable presumption that the claim limitation is to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites sufficient structure, material, or acts to entirely perform the recited function. Absence of the word “means” (or “step”) in a claim creates a rebuttable presumption that the claim limitation is not to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is not interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites function without reciting sufficient structure, material or acts to entirely perform the recited function. Claim limitations in this application that use the word “means” (or “step”) are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action. Conversely, claim limitations in this application that do not use the word “means” (or “step”) are not being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action. 07-30-06 This application includes one or more claim limitations that do not use the word “means,” but are nonetheless being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, because the claim limitation(s) uses a generic placeholder that is coupled with functional language without reciting sufficient structure to perform the recited function and the generic placeholder is not preceded by a structural modifier. Such claim limitation(s) is/are: " prompt construction unit " in Claims 1, 11, and 16 . Because this/these claim limitation(s) is/are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, it/they is/are being interpreted to cover the corresponding structure described in the specification as performing the claimed function, and equivalents thereof. If applicant does not intend to have this/these limitation(s) interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, applicant may: (1) amend the claim limitation(s) to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph (e.g., by reciting sufficient structure to perform the claimed function); or (2) present a sufficient showing that the claim limitation(s) recite(s) sufficient structure to perform the claimed function so as to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. Claim Rejections - 35 USC § 112 07-30-02 AIA The following is a quotation of 35 U.S.C. 112(b): (b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention. The following is a quotation of 35 U.S.C. 112 (pre-AIA), second paragraph: The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention. 07-34-01 Claims 1-20 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention. Claims 1, 11, and 16 recite the limitation " … the first content being associated with one or more output values of a statistical test … providing the one or more parameter values and the historical data to a calculation tool to generate one or more output values associated with the statistical test … the instruction string comprising instructions to the generative model to generate an interpretation of the one or more output values as the first content … " in lines 8-27; lines 4-23; and lines 5-24 respectively, which rendering these claims indefinite because (1) it is unclear whether the first two instances of " one or more output values " are the same or different; and (2) if they are different, which instance of " one or more output values " is referred by " the one or more output values " recited in the third instance. Clarification is required. Claims 2-10, 12-15, and 17-20 are rejected for fully incorporating the deficiency of their respective base claims. Claims 4, 14, and 19 recite the limitation " … providing the one or more assumed parameter values with the one or more parameter values to the calculation tool to generate the one or more output values " in lines 6-7; lines 4-5; and lines 5-6 respectively, which rendering these claims indefinite because " … the first content being associated with one or more output values of a statistical test … providing the one or more parameter values and the historical data to a calculation tool to generate one or more output values associated with the statistical test … the instruction string comprising instructions to the generative model to generate an interpretation of the one or more output values as the first content … " is also recited in their respective based claims and it is unclear which instance of " one or more output values " recited in their respective based claims is referred by the " one or more output values " recited here. Clarification is required. Claims 5, 15, and 20 recite the limitation " … determining a preferred output format for the user based on user data associated with the user … the instruction string comprises instructions to the generative model to determine a preferred output format for the user based on the user data and to present the interpretation in the preferred output format as the first content " in lines 4-10; lines 2-8; and lines 3-9 respectively, which rendering these claims indefinite because " … the instruction string comprising instructions to the generative model … " is also recited in their respective based claims, and (1) it is unclear whether the first two instances of " preferred output format " are the same or different; (2) if they are different, which instance of " preferred output format " is referred by " the preferred output format " recited in the third instance; and (3) it is unclear the instance of " instructions to the generative model " recited here is the same as or different to the instance of " instructions to the generative model " recited in their respective based claims. Clarification is required. Claim limitation “ prompt construction unit ” in Claims 1, 11, and 16 invokes 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. However, the written description fails to disclose the corresponding structure, material, or acts for performing the entire claimed function and to clearly link the structure, material, or acts to the function. ¶ [0076] with FIG. 3 of the specification describes that the prompt construction unit 124 includes a prompt formatting unit 302 and a prompt submission unit 304 and ¶ [0086] with FIG. 4 of the specification describes that t he prompt construction unit 124 includes a content retrieval unit 402 , and a format converting unit 404 . It is unclear which configuration (¶ [0076] with FIG. 3 or ¶ [0086] with FIG. 4) will be used to represent " prompt construction unit " of the claim. Also, no association between the structures described in FIGS. 6-7 and the functions performed by " prompt formatting unit 302 ", " prompt submission unit 304 ", " content retrieval unit 402 ", or " format converting unit 404 " can be found in the specification. Therefore, the claim is indefinite and is rejected under 35 U.S.C. 112(b) or pre-AIA 35 U.S.C. 112, second paragraph. 07-34-23 Applicant may: (a) Amend the claim so that the claim limitation will no longer be interpreted as a limitation under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph; (b) Amend the written description of the specification such that it expressly recites what structure, material, or acts perform the entire claimed function, without introducing any new matter (35 U.S.C. 132(a)); or (c) Amend the written description of the specification such that it clearly links the structure, material, or acts disclosed therein to the function recited in the claim, without introducing any new matter (35 U.S.C. 132(a)). If applicant is of the opinion that the written description of the specification already implicitly or inherently discloses the corresponding structure, material, or acts and clearly links them to the function so that one of ordinary skill in the art would recognize what structure, material, or acts perform the claimed function, applicant should clarify the record by either: (a) Amending the written description of the specification such that it expressly recites the corresponding structure, material, or acts for performing the claimed function and clearly links or associates the structure, material, or acts to the claimed function, without introducing any new matter (35 U.S.C. 132(a)); or (b) Stating on the record what the corresponding structure, material, or acts, which are implicitly or inherently set forth in the written description of the specification, perform the claimed function. For more information, see 37 CFR 1.75(d) and MPEP §§ 608.01(o) and 2181. Claims 2-10, 12-15, and 17-20 are rejected for fully incorporating the deficiency of their respective base claims. Claim Rejections - 35 USC § 102 07-06 AIA 15-10-15 In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status. 07-07-aia AIA 07-07 The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action: A person shall be entitled to a patent unless – 07-12-aia AIA (a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention. 07-15-03-aia AIA Claim s 1-20 are rejected under 35 U.S.C. 102(a)(2) as being anticipated by Linder et al. (US 2026/0037566 A1, priority date: 04/16/2023), hereinafter Linder . Independent Claims 1, 11, and 16 Linder discloses a data processing system (Linder, ¶¶ [0048]-[0049] with 100 in FIG. 1: a computer system 100 ¶ [0113] with 420 in FIG. 4: the data converter 404, the first configuration module 406, the second configuration module, the presentation module 414, and the request management module 418 may be hosted on a computing platform 420 a computing platform 420; the first inference engine 408 and the second inference engine 412 may also execute on the computing platform 420, or may be hosted remote from the computing platform) comprising: a processor (Linder, ¶¶ [0052]-[0053] with 112 in FIG. 1: processor 112); and a machine-readable storage medium (see ¶¶ [0126] and [0129] of the specification to exclude signal per se .) (Linder, ¶¶ [0052] and [0054]: the memory 114 may include any non-transitory computer readable medium) storing executable instructions that, when executed, cause the processor alone or in combination with other processors to perform operations of (Linder, ¶ [0053]: the processor 112 may be capable of processing instructions for execution within the computing device 110 or computer system 100; the processor 112 may be capable of processing instructions stored in the memory 114 or in the data store 118; ¶ [0054]: the memory 114 may store program instructions, program data, executables, and other software and data useful for controlling operation of the computing device 110, and for configuring the computing device 110 to perform functions for a use): receiving, via a user interface of a client device of a user (Linder, ¶¶ [0064], [0067], and [0071] with 206 and 212 in FIG. 2: the workstations may provide user interfaces for providing information to human users and receiving input from human users, along with any suitable combination of sensors, actuators, machinery, and the like for performing manufacturing tasks, along with one or more computing devices for integrating operation of the workstation hardware into a manufacturing process, providing instructions to a human user, receiving input or inquiries from the human user, and so forth; each client device 206 generally includes a user interface such as a graphical user interface, command line interface, or other user interface for user interaction with a facility 204 and other network resources; ¶ [0112] with 416 in FIG. 4: the user interface 416 may provide an interface for a user to configure a request, and to receive results from the request; ¶ [0140] with 600 in FIG.6: a user interface 600 for an application that controls a process in a manufacturing environment may include a number of features, such as a graphical display (e.g., of process inputs, outputs, video monitoring, time series data, etc.), a text display (e.g., of user instructions, quantitative data, tables, etc.), and a user input field (e.g., to receive text input, or input via other user controls such as checkboxes, radio buttons, drop down lists, and so forth)), a first request for first content to be generated by a generative model (Linder, ¶ [0073]-[0077] with 216 in FIG. 2: the other resources 216 may include an inference engine that uses, e.g., generative artificial intelligence (or generative AI) to create new content based on a corpus of training data; the term "generative AI," as used herein, refers to artificial intelligence technologies that have the capability to generate new content, including text, images, video, program code, and the like; some well-known examples of generative AI technologies include Generative Adversarial Networks (GANs) and Transformers; GANs use two neural networks, a generator and a discriminator, to concurrently produce and evaluate new content to achieve highly realistic outputs (relative to the training data); transformers have been successful in natural language processing tasks, and can generate coherent and contextually relevant text based on a prompt; an inference engine may be supported by a model for generative AI, such as a large language model trained on large amounts of natural language data, e.g., a large corpus of natural language documents containing text, to generate human-like responses to prompts; these large language models use machine learning techniques such as deep neural networks to learn patterns and structures in language data, which they can then use to generate new text that is similar in style and content to the training data; one well-known example of a large language model is OpenAI's GPT-3 (or GPT- 3.5, GPT-3.5 Turbo, GPT-4, GPT-4 Turbo), which has been trained on a massive corpus of text data and has achieved impressive performance on a wide range of natural language tasks; such a model may also be fine-tuned using specific user data (such as data specific to an industry or company using the model to generate analysis and recommendations), or a new large language model may be created specifically for a particular industrial or commercial application; generative AI models have been created for tasks such as generating images in response to natural language prompts, generating computer code in response to prompts, converting text into spoken audio, converting audio into text, converting text into numerical form, converting images into text, and so forth; the inference engine may also or instead employ other machine learning and/or analytical techniques to perform analysis and make recommendations based on manufacturing data and other available data sources; an inference engine based on generative AI can be used to create recommendations by analyzing patterns and similarities in user-provided datasets of behaviors or preferences, and then generating new recommendations based on that analysis), the first request including a first prompt describing the first content to be generated (Linder, ¶ [0112] with FIG. 4: the user interface may be coupled to a user request management module 418 that is configured to receive a user selection of data from the manufacturing environment 402, along with one or more user criteria for creating a prompt or other version of the request formatted for use by the first inference engine 408 and the second inference engine 412; the user request management module 418 may also be configured to receive a user selection of prompt characteristics for the prompts to the first inference engine 408 and/or second inference engine 412; the user interface 416 may also or instead present one or more pre-configured analysis requests for a user to select from, and/or may receive analysis descriptions as text input directly from a user), the first content being associated with one or more output values of a statistical test (Linder, ¶¶ [0095]-[0098] with 308 in FIG. 3: the request for analysis may request specific statistical analyses (mean, standard deviation, etc.), comparisons (e.g., compare operator 1 to operator 2, or manufacturing line 1 to manufacturing line 2), optimizations (e.g., "How do I make this faster?" or "How do I reduce error rates?"), or other analysis of result effective variables and the like; the request may include a request for a pie chart, graph, time series comparison, or other type of visual output based on the data input and any intervening analysis by the large language model; ¶¶ [0114]-[0115] with FIG. 5: a process controlled by individual applications; provide a useful framework for modeling execution timing by providing an initial, implicit model for workflow (based on application control logic, user interface structure, and so forth) that also facilitates automated detection of process steps based on execution flow, as well as detection and measurement of the contributions of individual functional blocks within an application to the process timing; by gathering data in this manner, an embedding or other model can be created to evaluate similarity of process steps and then descriptors for timing such as statistical descriptors can be associated with the embedding space based on individual timing data for each process step identified for each application; automating target cycle time predictions from a manufacturing step or application; a properly instrumented line can facilitate capture of relevant data including detected steps, application input, and execution graph transitions, that permit the creation of empirical models of process timing to more accurately reflect an implemented process based on statistical timing data or other descriptors for similar steps that make up the process; permitting the use of statistical distributions or other statistical descriptors derived from empirical timing data rather than a single, scalar timing value for each process step; ¶¶ [0134]-[0138] with 514, 516, and 518 in FIG. 5: in step 514, receiving a new application (also referred to herein as an "unknown application"), or any other process description suitable for use with the execution model; in step 516, estimating an execution time for the new application with the execution model; calculate a process time, or a range or distribution of process times representing an estimated execution time for the process controlled by the new application; in step 518, refining the execution time for the new application, e.g., by monitoring usage of the application after creating the initial execution time estimate; incrementally adjusting timing distributio n(s) for the new application as data is acquired, or regenerating the execution model with the updated data; in general, statistical tuning may be performed over any suitable interval; the execution time estimate may be periodically compared to actual execution timing, and an update or other review may be automatically recommended when the actual execution timing deviates in some manner (e.g., beyond a threshold for maximum excursion in individual times , maximum change in a mean or median , change in the standard deviation , and so forth) from the execution time estimated based on the model; a user interface configured to receive a submission of the application, including any of the application descriptions , from a user and to display the estimated process time for the application to the use; ¶ [0140]: user interface elements may be mapped into an embedding space using any suitable feature extraction, analysis, and the like; from the embedding space, time estimates or step time models may be extracted that are associated with each element of the user interface (e.g., f(a), f(b), f(c)), e.g., based on a most similar prior application , or based on a location within the embedding space; these may be statistical descriptors such as a mean , a median , a mode , a variance , a standard deviation , a range , and so forth, or these may be a distribution of time values associated with the location in the embedding space; ¶¶ [0164]-[0165] and [0178]: generating a test plan based on a video of a manufacturing process, which may usefully include narration by the engineer or other personnel while the activity is being recorded in order to provide a contemporaneous oral description; generating one or more test metrics for each of a plurality of segments in the video, and then creating a test plan based on the one or more test metrics ; e.g., where a validation test is identified in the video (assembly weights between x and y), the test metrics may include a test to confirm that the validation is performed (by a user, or automatically in the manufacturing process) and a result is recorded; this may be performed automatically in a sandbox or other test environment, or in a live manufacturing environment, e.g., with suitable safeguards and/or human monitoring; ¶¶ [0187]-[0191]: computer-assisted test plan creation using generative AI; validation is the process of establishing documentary evidence demonstrating that a procedure, process, or activity consistently produces the expected results or product quality; a test plan for validation outlines the approach, scope, resources, and schedule for validating a product, system, or process to ensure it meets specified requirements and standards; the test plan may include an explanation of what is being validated (product, system, process, etc.), a description of the boundaries of the validation process, a list of items or functionalities to be validated, an identification of any items or functionalities that are out of scope, and so forth; the test plan may also include objectives of the test plan, including clear, measurable validation objectives that align with project goals, along with examples include ensuring compliance with regulations , verifying functionality, assessing performance, etc.; the test plan may include a validation approach, such as a description of the overall approach to be taken for validation (e.g., black-box testing, white-box testing , etc.), an explanation of any specific methodologies or techniques to be used, validation criteria (including detailed criteria or standards against which the validation will be measured, and any regulatory requirements, customer specifications, or internal standards), test cases (including a comprehensive list of test cases designed to validate each requirement or functionality), a test environment (including hardware, software, tools, etc.), and any special configurations or setups required for testing ; the test plan may also specify, e.g., the assignment roles and responsibilities to team members involved in the validation process, a timeline for the validation process including milestones, testing phases, and deadlines, an allocation of time for each test case or testing activity, a list of resources required for validation including personnel, equipment, software licenses, etc., budget considerations, risk and mitigation strategies, and a plan for documenting test results, issues encountered, and any deviations from expected outcomes; by establishing and following a structured approach, a test plan for validation can help to ensure thorough testing and verification of a product, system, or process, leading to confidence in its quality and compliance with requirements; generative artificial intelligence can be applied to a manufacturing process to create a test plan related documents for validation, including proof of execution of various tests , process steps, and the like; a test plan for validation (also referred to herein as a "validation plan") may be automatically generated; the validation plan may be generated by a large language model or other specifically trained AI model, based on text description and any suitable prompts (e.g., requesting a specific type of analysis , specific regulatory or manufacturing context, specific type of testing , etc.) to generate a test plan; the test plan may be simulated or otherwise modeled for testing purposes; large language model may interrogate an application developer to parameterize the testing requirements ; the user interface for an application and/or administrative interface for a facility subject to the testing, may indicate validation status , validations in progress, validation test failures , and so forth); executing a query on one or more expert knowledge sources to obtain parameters associated with the statistical test, the parameters including one or more input parameters, one or more output parameters, or a combination thereof; constructing, based on at least a portion of the parameters using a prompt construction unit, one or more prompts to query the user for one or more parameter values associated with the parameters; receiving, via the user interface, one or more responses to the one or more prompts, the one or more responses including the one or more parameter values (Linder, ¶ [0069] with FIG. 2: the content sources 210 may include any sources of content for use herein; this may, e.g., include industry data, manufacturing specifications and standards, user guides, code repositories, device specifications, customer requirements, databases, collaborative resources, inventory or price data, and so forth; ABSTRACT and ¶ [0083]: data from manufacturing is highly uncontextualized and siloed, requiring expert knowledge of context and substantial data pre-processing to support meaningful queries and visualizations; data for a number of different sources in a manufacturing context/environment can be retrieved and converted into an intermediate representation in a natural language or near-natural language form, which can in tum be ingested by a generative AI engine, along with suitable prompts by the user to summarize, analyze, and make recommendations based on the data; ¶¶ [0073]-[0074]: generative AI technologies use machine learning models such as deep learning networks to analyze vast amounts of data, based upon which they can learn patterns, styles, or rules, or otherwise encode knowledge contained in the training data, and then use this training to generate new content, e.g., using transformers or other coder/decoders, that is similar to the training data; large language models have also been used in chatbots and virtual assistants , where they can understand and respond to human language input in a natural and conversational way; these models can encode and benefit from broad knowledge bases and expertise in various domains on which the model was trained, and may provide a natural language or other intuitive interface for user interactions; ¶¶ [0077]-[0079]: generative AI may be used with user-provided data and a large language model to provide summarization, outlining, information extraction, information expansion (e.g., prompted content creation) , recommendations, translation, rephrasing, sentiment analysis, coding, text to programmatic commands, code to natural language, natural language to database query (e.g., SQL), classification, grammar correction, conversational interaction (e.g., for chat bots or the like), and so forth; adapt inferencing based on machine learning for use in a manufacturing context; existing pre-trained models may be used sequentially or iteratively to prepare manufacturing data for use with existing foundation models, e.g. by initially requesting a data summary , and then presenting this summary along with a specific request for analysis or recommendations; ¶¶ [0083]-[0092] with 302 and 304 in FIG. 3: converting a range of data for a manufacturing process into a form that can be ingested by a generative AI engine, and providing suitable prompts to the generative AI engine to summarize, analyze, and make recommendations based on the data; in step 302, receiving data such as manufacturing data from a manufacturing system or other process, facility or the like; in step 304, converting manufacturing data into an intermediate representation; e.g., converting the manufacturing data into a natural language representation or near natural language representation including a human-readable description of the manufacturing data, where a natural language representation would include language closely mimicking human language, a near natural language representation permits the use of more encoded data in text form, e.g., using schemas, fields, or the like; e.g., structured queries, Boolean operators, wildcards, and other logical operators, program command syntaxes, and the like, may be incorporated into a text-based near natural language representation to provide greater flexibility in the format of data inputs; the manufacturing data may also or instead include process data from at least one sensor in the manufacturing process, such as a sensor controlled by the application; this may include sensor data such as discrete or time series measurements of, e.g., temperature, weight, speed, force, voltage, strain, and so forth; sensor data may be raw sensor data, filtered sensor data, processed sensor data (e.g., via descriptive statistics or the like), and so forth; the intermediate representation may be any natural language characterization or near natural language characterization that allows an inference engine to use a pre-trained large language model (such as ChatGPT 3, ChatGPT 3.5, ChatGPT 4, Google's Bidirectional Encoder Representations from Transformers (BERT), and so forth) or the like to process or reason about the data semantically, bringing in knowledge from across a wide domain of training materials; the inference engine may also be finetuned on data from a specific domain , such as a particular industry, academic or scientific discipline, and so forth; ¶¶ [0093]-[0094] with 306 in FIG. 3: in step 306, requesting a summarization of the (near) natural language representation from a first model, such as by requesting a summarization of a human-readable description from a first large language model; the first model may more generally include any suitable foundation model, such as a language model or a large language model pre-trained on a large corpus of text; the model may also or instead include a refined large language model, or a language model trained on, e.g., domain-specific content for a manufacturing environment or the like; the model may be also or instead include a student model trained with any of the above, or any other compressed model or the like suitable for, e.g., edge computing, event processing, or other deployment in a local context; facilitating an automated conversion of the intermediate representation into a natural language description using the processing power and knowledge encoded in the large language model; in order to request a summarization, the intermediate representation of the manufacturing data may be presented to the model, and the model may be requested, e.g., with a suitable prompt or the like, to generate a description or summarization of the intermediate representation; the request for a summarization may also be further parameterized using natural language or near natural language to manage a prompt to a language model or the like, e.g., to request specific analysis or recommendations, or to focus the analysis on particular areas of interest; e.g., the input to the model may request initial insights relating to, e.g., possible data anomalies or result effective variables , or may provide instructions about how to organize or summarize the data; ¶¶ [0095]-[0098] with 308 in FIG. 3: in step 308, requesting an analysis of the summarization; presenting the summarization from the first model to a second model, along with a request for an analysis; the request for analysis may be presented as a prompt to a language model, or as any other suitable request formatted for the model as appropriate; the request may include a request for one or more result effective parameters for the manufacturing process, e.g., variables that affect the speed, cost, quality, or other characteristic(s) of the process; the request may include a request for possible improvements, quality control measures, modifications, and so forth; a user can also ask for recommendations to improve the process, or for recommendations on suitable database queries, code revisions, and so forth; these latter recommendations may be parsed back to SQL, user interface reconfigurations, application code, or the like by the model, or may be presented in text form for use by a user that receives a corresponding result; requesting a summarization (of near natural language description) from a first model (such as a first language model); one of the models may be a different model, e.g., a language model trained or fine-tuned with domain-specific training data for the manufacturing environment or some other knowledge domain ; e.g., by refining the first model for the current manufacturing environment, a summarization can be rendered that more accurately expresses the operating environment for downstream use by non-domain-specific inference engines; the request may include a request for a pie chart, graph, time series comparison, or other type of visual output based on the data input and any intervening analysis by the large language model; ¶¶ [0102]-[0107] with FIG. 4: the manufacturing environment 402 may generate manufacturing data from, e.g., sensors, human inputs, applications, and the like that are associated with a manufacturing process as a data stream or as raw data; this data may be filtered, augmented, aggregated, or otherwise processed for storage and/or for subsequent use in generating recommendations; manufacturing data may also or instead include external data such as process documentation, user manuals, third party data, customer requirements, product specifications, regulations, industry standards, inventory and price data, process images, process videos, and so forth; this may be converted into text form for use in natural language processing, e.g., by converting numbers to text, or providing descriptors for data tables, time series data, and so forth; sensor metadata may include context for a sensor that describes, e.g., intrinsic properties of the sensor (units of measurement, range of measurement, accuracy, sampling rates, etc.) or information about how the sensor fits into a manufacturing process (e.g., location, purpose, connected applications, and so forth); the data converter 404 may receive the manufacturing data from the manufacturing environment 402 and generate a near natural language description of the manufacturing data for use by a language model; the data converter 404 may apply scaling strategies to manage the quantity and quality of data that is presented to inference engines, which may include filtering real time data to a time scale of interest, and labeling data according to a type of event, source of event, and so forth, in order to compress the data that requires processing; the data converter 404 may extract statistical descriptions (minimum, maximum, range, mean , median , mode , variance , trend analysis, regression parameters , etc.); the data converter 404 may hierarchically or iteratively apply a language model, e.g., to summarize groups of events in a related meta-segment of data, and then to summarize a group of such summaries, in order to distill the raw data before initiating an analysis; a first configuration module 406 may be used to present the converted data from the manufacturing environment 402 to a first inference engine 408, such as a language model, large language model, refined language model, student model, or the like; a language model such as a large language model may be used to advantageously support inferences based on a knowledge domain derived from a large corpus of text-based documents upon which it was trained; the first configuration module 406 may be configured to request a natural language summary of the data from the manufacturing environment 402, e.g., by presenting the near natural language description from the data converter 404 to a language model or other first inference engine 408 along with suitable prompts; ¶ [0112] with FIG. 4: The user request management module 418 may provide a user-friendly interface for creating requests on one hand, while managing the first configuration module 406 and the second configuration module 410 so that suitable data and prompts (or the like) are presented to the inference engines to generate a response; e.g., the user interface 416 may present available data sources from the manufacturing environment 402 as check boxes, drop down lists, or the like so that a user can select what data is to be used when performing an analysis; the user request management module 418 may receive the various parameters from a user, and may use these parameters to configure the first configuration module 406 and the second configuration module 410 to generate suitable, corresponding prompts for use by the inference engines in generating a response to the request received from the user; ¶¶ [0116]-[0126] with FIG. 5: in step 502, identifying an application, such as an application that controls a manufacturing process; in step 506, segmenting the application into steps for purposes of characterizing an execution time for the application; programmatically inferring process steps and creating the execution graph based on an analysis of one or more of the software modules in an application; programmatically inferring a step based on user inputs to the software module, or based on features or functions of a user interface for the application; e.g., where user interaction is requested by a widget or the like, e.g., to initiate an operation, to indicate approval of a result, to request an inspection, and so forth, these programmatic demarcations may be identified and associated with steps in the process; create an embedding, feature vectors, parameters, or the like for evaluating similarity to other steps to facilitate analysis of a new, unknown application; other context (e.g., triggers for actions within the application, variables maintained and updated by the application, records captured by the application, machines and/or devices controlled by or monitored by the application, connectors to other applications and process entities, and so forth) may also be used to infer segmentation and identify process steps; each interaction point in an execution graph or other data structure describing these elements provides a potential demarcation for a process step that may be modeled for step timing; the steps in the application, or in the process controlled by the application, may include one or more steps controlled directly by the application, e.g., where the application autonomously executes a process step based on timing, sensor feedback, trigger events, and so forth; the steps may also or instead include one or more steps controlled by a user of the application, e.g., where the user interface prompts a user for input during the process; a preliminary time estimate for each step can also help to converge more quickly and efficiently at an accurate characterization of the process or help to identify errors in segmentation; an initial timing expectation for an application, or for one of the steps, may be obtained from a variety of sources; the timing expectation may be based on manual user steps in the process, prior observations, design expectations , regulations or industry standards applicable to the process, and so forth; in step 508, monitoring execution of the application; applications for a manufacturing process will run in real time, and data may be collected in substantially real time for any monitored aspects of the process; the realized timings for each step may be collected over any suitable window of time, and at any suitable temporal resolution, and may be based on any monitored inputs such as sensor feedback, user input to an application prompt , machine vision, and so forth; thus, monitoring may include acquiring timing data such as a timing data distribution for each step, and/or one or more other statistical descriptors or other quantitative descriptions; in step 510, determining whether there are additional applications to be analyzed; if there are additional applications, return to step 502, where the next application is identified and processed (e.g., for segmentation and monitoring); if there are no additional applications, proceed to step 512 where an execution model can be created; ¶ [0129]: the user interface (represented as screen shots, images, user interface code, or the like) may provide a basis for identifying steps and/or creating an embedding in a manner that facilitates step time estimation for a new process; the regression parameters for the regression model may also be selected based on the embedding, e.g., by selecting regression parameters based on similarity to other applications (or other process descriptions); ¶ [0133]: permits the creation of company-specific, industry-specific, or process-specific databases of step time models for expected execution timing; certain user interface elements, widgets, or functions may be associated with particular execution models, and corresponding step timings; similarly, a particular type of step (e.g., insert four screws with a T9 Torx™ screwdriver) may have similar timing expectations independent of the workpiece or manufacturing context; certain types of processes have steps prescribed by regulation, industry standard, or common practice, and may usefully be modeled across companies and/or processes; e.g., a safety audit, pharmaceutical line clearance, or visual part inspection may have similar or identical use cases across users; different application categories may be used to estimate timing; e.g., an application may be categorized as "assembly" (simple, moderate, complex, ... ), machining, quality control, etc., and this may be used to establish an initial timing estimate for a particular application, or to parameterize process steps within a latent space or embedding, which may also include mixed-type categories (e.g., 80% assembly, 20% quality control), each of which may receive a separate execution model, or separate embeddings when evaluating an unknown application for purposes of estimating execution timing; these characteristics may be determined based on descriptive metadata, inferred from characteristics of an application, or using any of the other techniques; ¶ [0139]: the user interface may facilitate use of the execution model 522 by permitting a user to provide or identify a new application , and to apply the execution model 522 to the new application for derivation of an execution time estimate ; the user interface may also facilitate subsequent steps such as estimate validation , estimate refinement , and so forth, as well as preceding steps such as creation and tuning of the execution model 522; ¶ [0156]: receiving a new application, such as an unknown application that has yet to be classified); retrieving historical data including historical parameter values of the parameters, historical data records requested by the user, or a combination thereof associated with the statistical test; providing the one or more parameter values and the historical data to a calculation tool to generate one or more output values associated with the statistical test (Linder, ¶ [0077]; generative AI can be used to analyze patterns in user behavior data to generate new recommendations based on what users have done in the past ; generative AI can employ content-based filtering, which involves analyzing the characteristics of items to make recommendations that a user may find helpful; generative AI can be used to analyze manufacturing data, user actions, and so forth to generate new recommendations for operating a manufacturing line or the like based on similarities to other manufacturing processes; a hybrid approach may also be used, where recommendations combine collaborative filtering and content-based filtering to make recommendations; ¶¶ [0095]-[0098] with 308 in FIG. 3: the request for analysis may include one or more analysis parameters presented in natural language to the second model; after obtaining a summarization in natural language form from the generative AI system, a second model can be used to analyze or reason about the summarization, or about multiple summarizations concurrently; given a stack of summarizations of an execution of an application (" Alice completed app A in an average of 30 seconds with an average of 5 defects per day, Bob completed app A in an average of 36 seconds with an average of 3 defects per day, ... "), a request to the model for analysis may be stated as " Who is the best operator on app A in terms of time?" or " Who is the best operator in terms of defects? "; requesting an analysis of the summarization from a second model (such as a second language model); by refining the second model, analysis and recommendations may be more closely tied to that operating environment; the request may include a request for a pie chart, graph, time series comparison, or other type of visual output based on the data input and any intervening analysis by the large language model; ¶¶ [0099]-[0101] with 310-312 in FIG.3: in step 310, parsing one or more recommendations from the analysis; parsing the results into a different format or adapting the results to a specific use context; parsing recommendations may include manual/human handling of output from the second model, or this may include an explicit request for code, process step descriptions, or the like from the second model as part of the recommendations; a third model may be used to convert the recommendations from the second model into computer executable code, e.g., after review by a human technician, administrator, or the like; receiving the analysis from the second model and presenting the analysis to a user; ¶¶ [0108]-[0109] with FIG. 4: a second configuration module 410 may be used to present the output from the first inference engine 408 to a second inference engine 412; e.g., the second configuration module 410 may be configured to receive a natural language summary (e.g., of manufacturing data from the manufacturing environment 402), and to present a request based on the natural language summary to a second large language model; more specifically, the second configuration module 410 may present the natural language summary to a second language model or other second inference engine 412, along with a request for a recommendation, e.g., in a natural language prompt or other request format adapted to the inference engine; ¶ [0115]: using applications as a framework for modeling execution timing provides numerous advantages including (a) providing an initial, implicit model for workflow based on the application control logic, (b) facilitating automated detection of process steps controlled by each application (e.g., based on static or dynamic analysis of application execution flow), (c) permitting feature-based comparison to portions of a new or unknown application to select suitably similar timing models , (d) facilitating detection of the contributions of individual processing routines, steps, and the like to the process timing e.g., by monitoring an application during execution, and (e) permitting the use of statistical distributions or other statistical descriptors derived from empirical timing data rather than a single, scalar timing value for each process step; ¶¶ [0124]-[0125]: it is possible to proceed directly to monitoring and statistical modeling without initial estimates of execution times for individual steps, or for the overall process; the timing expectation may be based on manual user steps in the process, prior observations , design expectations, regulations or industry standards applicable to the process, and so forth; the timing data may be stored for an application, e.g., as raw timing data, as a distribution , as statistical descriptors (e.g., mean, median, variance, and so forth), or in any other suitable format for executing time modeling ; ¶¶ [0127]-[0133] with FIG. 5: in step 512, creating an execution model for the application, which includes the generation of an embedding, latent space, or other feature-based representation of each process step that permits a similarity analysis for steps of a new, unknown application, along with a corresponding timing model such as an average , a range , a variance , a distribution , and so forth, as well as combinations of the foregoing; the step timing models may include distributions (e.g., Normal/Gaussian distributions) for the timing interval for each step, rather than a single numerical value, and/or any other suitable statistical descriptors or the like; as a significant advantage, where execution steps are modeled as distributions , actual execution times for individual steps that are captured during execution may be reported relative to confidence intervals or the like rather than as binary evaluations of whether a particular timing metric was met, and multiple consecutive steps can be better modeled for overall expected process timing; an application for a manufacturing process may include three widgets (or other modular software components or the like) that provide a first button to perform a task, a second button to approve a step/item, and a third button to request assistance; as an initial pass, each of these items may have been assigned an expected time based on a contribution of the associated task to the total step time; after collecting real time data, the overall step may be found to take more or less time than initially estimated; this time differential may be allocated among the widgets, each representing a normally distributed subpopulation of the total step time, e.g., by fitting the data to a Gaussian Mixture Model or similar distribution in a supervised machine learning process in order to estimate parameters for each step; more generally, any technique for aggregating a number of probability distributions may be used to estimate an aggregated timing for a step, or for an application that performs a number of steps, in order to support the creation of an empirical execution model as contemplated herein; a regression model or the like may be used to associate the contribution of steps, e.g., concurrent, overlapping, or alternative steps, to an overall process time; creating the execution model may include creating a regression model including one or more parameters relating step times or other outputs of the embedding for the application to a process time for an application; regression parameters may be calculated for a linear regression or similar model that estimates total process time based on the sum of times for process steps, more specifically based on the execution model for the underlying process; the step time for a new application may usefully be estimated by generatively estimating step times or distributions with the embedding based on characteristics of an application, and then combining these step times using a regression model to estimate the overall process time for the new application; matrix factorization may be used to model the contribution of each step, widget or the like to the total execution time, which is a mathematical technique that decomposes a matrix into a product of two or more simpler matrices; the goal of matrix factorization is typically to find a low-rank representation of the original matrix that captures the underlying structure and patterns in the data, thus reducing the dimensionality of the original matrix to facilitate analysis and processing of matrix data, which can also often help to characterize the underlying structure and pattern of data in the matrix, e.g., for use in generating executing timing estimates ; in the context of machine learning solutions to execution time modeling , matrix factorization may be used to analyze raw timing data to create a distributed , mixed model or the like for a particular application; e.g., a matrix may be created including a latent space of step types (e.g., image widget, text widget, other step types) as columns, and actual execution times in rows; creating an execution model may include creating an embedding for components of an application; the choice a particular technique may depend on the nature of the data (e.g., tokenization for text, or resizing/normalization for images) and a variety of other factors; the embedded space can usefully represent steps in the process in a manner that facilitates identification of similar steps for purposes of selecting a suitable step timing model or step timing estimate; other techniques may also or instead be used; e.g., where steps are sequential and/or have a dependency, a recurrent network or attention-based model or transformer may usefully be employed to draw inferences about total process time; ¶¶ [0135]-[0138] with 516, 518, 522, and 524 in FIG. 5: in step 516, estimating an execution time for the new application with the execution model, which may include decomposing, segmenting, or otherwise analyzing the application as described above, in order to identify process steps and transform the process steps into the embedded space to select a suitable timing model , e.g., based on similarity to other process steps contained in the execution model 522; each similar step (or set of steps) in the embedded space may then be used to provide step time estimates, and, where available, regression parameters for the associated process steps, which can be used in turn to calculate a process time, or a range or distribution of process times representing an estimated execution time for the process controlled by the new application; in step 518, refining the execution time for the new application, e.g., by monitoring usage of the application after creating the initial execution time estimate; the estimate for the application (and/or the execution model for all processes) can be updated continuously or periodically on a rolling basis, based on a predetermined historical window of execution data (e.g., preceding seven days, preceding four weeks, etc.); receive a new application, and to parse the new application into a second plurality of software modules; then map the second plurality of software modules to the first plurality of software modules, e.g., using an embedding; apply the execution model 522 to derive statistical descriptors for one or more steps in the new application based on proximity to one or more of the first plurality of software modules in an embedded space of the embedding; then determine an estimated process time for the new application, optionally using a regression model or the like to model alternative or overlapping process steps; ¶¶ [0140]-[0142] with FIG. 6: in order to estimate an execution time, the individual time estimates may be summed into an execution time ("Time") associated with the user interface 600; a total time for an application may be a sum of estimated times for individual steps, such as process steps inferred from the user interface and/or from other data; ¶ [0144]: submitting the application to a large language model, along with a prompt to generate a text-based summary of the application; the prompt may specify additional details or parameters for the summary, such as a desired length of the summary, format of the summary, informational content for the summary, and so forth; ¶ [0165]: mistakes may be, e.g., known failure points based on historical data , anticipated failure points for a new process, actual failure points observed during process creation, or any other predicted, likely, or highly undesirable failure points that might usefully be anticipated and managed while creating a process flow for an application used to control or perform the process); constructing a second prompt by the prompt construction unit as an input to the generative model, the prompt construction unit constructing the second prompt by appending at least the one or more prompts and the one or more responses with an instruction string, the instruction string comprising instructions to the generative model to generate an interpretation of the one or more output values as the first content (Linder, ¶ [0077]: generative AI can be used to analyze user behavior data, instrumentation or other process data, productivity and results, and to generate new recommendations, tools, analysis, and the like based in response to input data and any accompanying prompts , requests , analysis parameters , or the like; ¶¶ [0080]-[0082]: support a multi-model head end for a generative AI system; by fusing these heterogenous data types into a unified representation, a suitable latent space can then be configured for creating a decoder or generator network that can create new application code based on multi-modal process descriptions; a variety of prompt engineering and other augmentation techniques may be used to improve results from a foundation model that was not trained with a domain specific data set; e.g., retrieving current (i.e., non-training) data or other descriptive text and the like from a manufacturing process to augment a request from an inference engine, and/or augmenting model outputs by retrieving process-specific data relevant to one or more portions of the generative output; constrain a generative AI analysis or coding task to a particular manufacturing domain or process in order to facilitate the creation of useful results without requiring a full custom foundation model for, e.g., coding manufacturing applications; ¶¶ [0092]-[0094] with FIG. 3: while the method 300 shown in FIG. 3 focuses on text-based natural language processing, similar generative AI models may be created for other modes of input and/or output such as audio, speech, video, images, and so forth; the intermediate representation may usefully include any of the foregoing, provided the corresponding generative AI system can interpret the data; extracting domain-specific features of interest to augment a prompt or other request to an inference engine that uses a more generic foundation model; supplemental information may be explicitly provided by a user, or may be automatically generated using a domain-specific embedding, retrieval-augmented generation (RAG), or other techniques to automatically generate (or locate) and append supplemental information to the intermediate representation; ¶¶ [0096]-[0098]: requesting the analysis may include requesting computer readable instructions implementing one or more recommendations contained in the analysis; e.g., the input to the generative AI model may include a request to identify the operator with the lowest error rate, and recommend code changes in an application based on the behavior of that operator; many current large language models can generate executable code based on task descriptions or other specifications, and can be specifically requested to produce code, e.g., in a particular programming language or for a particular interpreter or environment; more refined models may also or instead be used to generate code suitable for a particular manufacturing context based on, e.g., available libraries, connectors, resources, code bases, and so forth; ¶ [0101] with 312 in FIG. 3: in step 312, performing further processing on the analysis; post-processing the output for presentation, e.g., for normalization, for consistency with terminology used in the manufacturing process, to format or organize the output data, to augment the results, and so forth; presenting recommended coding changes, along with an explanation for the nature of, and reasons for, the recommended changes; ¶ [0104] with FIG. 4: descriptive metadata may be used to create natural language characterizations of applications, application data, application logs, sensors, sensor data, and so forth, so that the data can analyzed within the context of the manufacturing process, which permits an inference engine such as a large language model to analyze and interpret manufacturing data in the correct context , rather than simply as a collection of structured or unstructured quantitative data; in particular, the metadata for the sensor(s) can impart physical meaning to quantitative data , and metadata for the application(s) can permit inferences about, or reasoning based on, the intended use and nature of the sensor data; a variety of techniques for automated data augmentation may be used to augment raw sensor/application outputs with suitable metadata to assist in interpretation for the purposes described herein; ¶¶ [0110]-[0111] with FIG. 4: a presentation module 414 may receive the recommendation from the second inference engine 412, and may perform any supplemental post-processing useful for presenting the results to a user; e.g., the presentation module 414 may receive a natural language recommendation from a second language model, and may parse the recommendation for presentation in a user interface 416 according to one or more user criteria; this may include filtering the recommendation, formatting the recommendation for presentation in a user interface 416 (e.g., as one or more windows showing relevant data, analysis, recommendations, code implementing recommendations, and so forth); e.g., language in the recommendation may be mapped to vocabulary, process descriptions, and the like from process documentation so that the results are cast in a rubric that can be readily understood and interpreted by a human reviewer who is familiar with the manufacturing environment 402; the presentation module 414 may usefully manage queries to the data store 403, e.g., to access raw data from the manufacturing environment 402 to augment analysis and recommendations that are output by the second inference engine 412, which permits inference based on a reduced-size, abstracted data set, while permitting augmentation with more granular data once a particular recommendation is identified; ¶ [0130]: matrix factorization may then be performed to identify the latent factors that explain variation in execution times across the different types of steps; these latent factors may then be used to predict a total execution time for an application (or other unit of process) based on constituent elements; ¶¶ [0151]-[0152]: text description of an application may be augmented with supplemental information to support multi-modal classification; ¶¶ [0157]-[0162] with FIG. 7: converting the new application into a text-based format or other natural language description; receiving a supplemental description of the new application which include any of the supplemental or augmented descriptions or data matched to the type of supplemental data used to train the classification engine 722; generating an embedding for the supplemental data to extract features, and/or transform the supplemental description for the new application into a vector or set of vectors that can be used as inputs for classification; classifying the new application by applying the classification engine to the new application, e.g., based on the description and the supplemental description, and/or any associated embeddings; the output of the classification engine 722 may be an application type, such as any type or category useful for managing applications in a manufacturing environment or the like; the type may also or instead be an industry category; the categories may include any category or group of functional categories, user categories, process categories, industry categories, and so forth useful for organizing and managing software in a manufacturing environment; produce an accuracy estimate such as a probability of belonging to one or more categories; a classification may also be used for any of the other application analysis or management tools, e.g., where a classification is used as supplemental information for input to another generative AI process; ¶ [0189]: in order to create a validation plan, an application may be transformed from an internal, executable representation into a text description, e.g., so that it can be compared to validation requirements; this text representation may be augmented with classification information, which may be derived from application metadata or other descriptive information, or derived by classifying the application based on the content and context of the application; the textualization may also or instead be augmented with images, video, and other multimedia or non-text content contained in the application, or with natural language descriptions of any of this content obtained, e.g., from suitable embeddings and/or inference engines; ¶ [0193]: permits the description to be augmented with additional data about the type or context of the application that are inferred from characteristics of the application, e.g., so that appropriate, corresponding testing or validation requirements can be identified and applied; ¶ [0195]: a wide range of supplemental data may be provided and used to augment application descriptions on one hand, and create an embedding used to characterize applications on the other; supplemental data may be acquired from a manufacturing environment such as individual user tendencies observed during monitoring, or user characteristics such as age, experience, efficiency, physical size, and so forth; the supplemental data may include physical characteristics of a workspace, such as a working volume, dimensions of a desk, locations of bins, screens, input/output devices, and so forth; these extra modalities, which may be provided as physical specifications, images, text descriptions, and the like, may be used to create a multimodal representation of an application or process that extends well beyond code and machinery, and can improve an embedding space used to extract relevant features from existing testing/validation requirements and to create testing or validation plans; ¶ [0200]: where new code or coding modifications are included in the test plan, these tools may be used, e.g., by applying natural language processing techniques or the like to interpret a description of a process, applying inference engines such as large language models to identify key components of the description (such as variables, functions, conditions, loops, etc.) that need to be converted into code, and then mapping these components to programming constructs); providing the first content to the client device; and causing the user interface to present the first content (Linder, ¶ [0101] with 312 in FIG. 3: receiving the analysis from the second model and presenting the analysis to a user; displaying the output from the second model (and/or third model, where code generation is separately performed); recording results of the analysis, and/or transmitting the results to one or more users; ¶¶ [0110]-[0112] with FIG. 4: presenting the results to a user; receive a natural language recommendation, and parse the recommendation for presentation in a user interface 416 (e.g., as one or more windows showing relevant data, analysis, recommendations, code implementing recommendations, and so forth); the user interface 416 may provide an interface to receive results from the request; ¶¶ [0137] and [0139]: a user interface configured to display the estimated process/step time for the application to the user; ¶¶ [0161]-[0162]: displaying the classification to a user in a user interface, e.g., for review or further action, or applying a policy or management rule for an enterprise to the new application based on the classification (e.g., to submit for review, to add to a repository, to limit authorized users, etc.); ). Linder further discloses a non-transitory computer readable medium (Linder, ¶¶ [0052] and [0054]: the memory 114 may include any non-transitory computer readable medium) on which are stored instructions that, when executed, cause a programmable device to perform functions of the method described above (Linder, ¶ [0053]: the processor 112 may be capable of processing instructions for execution within the computing device 110 or computer system 100; the processor 112 may be capable of processing instructions stored in the memory 114 or in the data store 118; ¶ [0054]: the memory 114 may store program instructions, program data, executables, and other software and data useful for controlling operation of the computing device 110, and for configuring the computing device 110 to perform functions for a use) Claims 2, 12, and 17 Linder discloses all the elements as stated in Claim 1 and further discloses selecting the calculation tool among a plurality of calculation tools based on the statistical test (Linder, ¶¶ [0095]-[0098] with 308 in FIG. 3: the request for analysis may request specific statistical analyses (mean, standard deviation, etc.), comparisons (e.g., compare operator 1 to operator 2, or manufacturing line 1 to manufacturing line 2), optimizations (e.g., "How do I make this faster?" or "How do I reduce error rates?"), or other analysis of result effective variables and the like; the request may include a request for a pie chart, graph, time series comparison, or other type of visual output based on the data input and any intervening analysis by the large language model; ¶¶ [0114]-[0115] with FIG. 5: evaluate similarity of process steps and then descriptors for timing such as statistical descriptors based on individual timing data for each process step identified for each application; automating target cycle time predictions permit the creation of empirical models of process timing to more accurately reflect an implemented process based on statistical timing data or other descriptors for similar steps that make up the process; permitting the use of statistical distributions or other statistical descriptors derived from empirical timing data rather than a single, scalar timing value for each process step; ¶¶ [0135]-[0138] with 516 and 518 in FIG. 5: in in step 516, estimating an execution time for the new application with the execution model; calculate a process time, or a range or distribution of process times representing an estimated execution time for the process controlled by the new application; in step 518, refining the execution time for the new application; incrementally adjusting timing distributio n(s) for the new application as data is acquired, or regenerating the execution model with the updated data; in general, statistical tuning may be performed over any suitable interval; the execution time estimate may be periodically compared to actual execution timing, and an update or other review may be automatically recommended when the actual execution timing deviates in some manner (e.g., beyond a threshold for maximum excursion in individual times , maximum change in a mean or median , change in the standard deviation , and so forth) from the execution time estimated based on the model; ¶ [0140]: user interface elements may be mapped into an embedding space using any suitable feature extraction, analysis, and the like; from the embedding space, time estimates or step time models may be extracted that are associated with each element of the user interface (e.g., f(a), f(b), f(c)), e.g., based on a most similar prior application, or based on a location within the embedding space; these may be statistical descriptors such as a mean , a median , a mode , a variance , a standard deviation , a range , and so forth, or these may be a distribution of time values associated with the location in the embedding space; ¶¶ [0164]-[0165] and [0178]: generating a test plan based on a video of a manufacturing process, which may usefully include narration by the engineer or other personnel while the activity is being recorded in order to provide a contemporaneous oral description; generating one or more test metrics for each of a plurality of segments in the video, and then creating a test plan based on the one or more test metrics ; e.g., where a validation test is identified in the video (assembly weights between x and y), the test metrics may include a test to confirm that the validation is performed (by a user, or automatically in the manufacturing process) and a result is recorded; this may be performed automatically in a sandbox or other test environment, or in a live manufacturing environment, e.g., with suitable safeguards and/or human monitoring; ¶¶ [0187]-[0191]: computer-assisted test plan creation using generative AI; validation is the process of establishing documentary evidence demonstrating that a procedure, process, or activity consistently produces the expected results or product quality; a test plan for validation outlines the approach, scope, resources, and schedule for validating a product, system, or process to ensure it meets specified requirements and standards; the test plan may include an explanation of what is being validated (product, system, process, etc.), a description of the boundaries of the validation process, a list of items or functionalities to be validated, an identification of any items or functionalities that are out of scope, and so forth; the test plan may also include objectives of the test plan, including clear, measurable validation objectives that align with project goals, along with examples include ensuring compliance with regulations , verifying functionality, assessing performance, etc.; the test plan may include a validation approach, such as a description of the overall approach to be taken for validation (e.g., black-box testing, white-box testing , etc.), an explanation of any specific methodologies or techniques to be used, validation criteria (including detailed criteria or standards against which the validation will be measured, and any regulatory requirements, customer specifications, or internal standards), test cases (including a comprehensive list of test cases designed to validate each requirement or functionality), a test environment (including hardware, software, tools, etc.), and any special configurations or setups required for testing ; the test plan may also specify, e.g., the assignment roles and responsibilities to team members involved in the validation process, a timeline for the validation process including milestones, testing phases, and deadlines, an allocation of time for each test case or testing activity, a list of resources required for validation including personnel, equipment, software licenses, etc., budget considerations, risk and mitigation strategies, and a plan for documenting test results, issues encountered, and any deviations from expected outcomes; by establishing and following a structured approach, a test plan for validation can help to ensure thorough testing and verification of a product, system, or process, leading to confidence in its quality and compliance with requirements; generative artificial intelligence can be applied to a manufacturing process to create a test plan related documents for validation, including proof of execution of various tests , process steps, and the like; a test plan for validation (also referred to herein as a "validation plan") may be automatically generated; the validation plan may be generated by a large language model or other specifically trained AI model, based on text description and any suitable prompts (e.g., requesting a specific type of analysis , specific regulatory or manufacturing context, specific type of testing , etc.) to generate a test plan; the test plan may be simulated or otherwise modeled for testing purposes; large language model may interrogate an application developer to parameterize the testing requirements ; the user interface for an application and/or administrative interface for a facility subject to the testing, may indicate validation status , validations in progress, validation test failures , and so forth). Claims 3, 13, and 18 Linder discloses all the elements as stated in Claim 1 and further discloses selecting the generative model among a plurality of generative models based on the statistical test (Linder, ¶¶ [0005] and [0097] with FIG. 3: requesting a summarization of the human-readable description from a first large language model; and presenting the summarization from the first large language model to a second large language model along with a request for an analysis including an identification of one or more result effective parameters for the manufacturing process; the second model and the first model may be the same (e.g., a single large language model used at both steps in the method 300, either concurrently or sequentially), or one of the models may be a different model, e.g., a language model trained or fine-tuned with domain-specific training data for the manufacturing environment or some other knowledge domain; ¶¶ [0107]-[0109] with FIG. 4: the first configuration module 406 may be configured to request a natural language summary of the data from the manufacturing environment 402, e.g., by presenting the near natural language description from the data converter 404 to a language model or other first inference engine 408 along with suitable prompts; the second configuration module 410 may be configured to receive a natural language summary (e.g., of manufacturing data from the manufacturing environment 402), and to present a request based on the natural language summary to a second large language model; i.e., the second configuration module 410 may present the natural language summary to a second language model or other second inference engine 412, along with a request for a recommendation, e.g., in a natural language prompt or other request format adapted to the inference engine; the first inference engine 408 or the second inference engine 412 may be a different inference engines, such as a language model trained or refined using a knowledge base for the manufacturing environment 402, or some other data set specific to the data context or the requested analysis). Claims 4, 14, and 19 Linder discloses all the elements as stated in Claim 1 and further discloses executing another query on the one or more expert knowledge sources to obtain one or more assumed parameter values associated with the parameters; and providing the one or more assumed parameter values with the one or more parameter values to the calculation tool to generate the one or more output values (Linder, ¶ [0069] with FIG. 2: the content sources 210 may include any sources of content for use herein; this may, e.g., include industry data, manufacturing specifications and standards, user guides, code repositories, device specifications, customer requirements, databases, collaborative resources, inventory or price data, and so forth; ABSTRACT and ¶ [0083]: data from manufacturing is highly uncontextualized and siloed, requiring expert knowledge of context and substantial data pre-processing to support meaningful queries and visualizations; data for a number of different sources in a manufacturing context/environment can be retrieved and converted into an intermediate representation in a natural language or near-natural language form, which can in tum be ingested by a generative AI engine, along with suitable prompts by the user to summarize, analyze, and make recommendations based on the data; ¶¶ [0073]-[0074]: generative AI technologies use machine learning models such as deep learning networks to analyze vast amounts of data, based upon which they can learn patterns, styles, or rules, or otherwise encode knowledge contained in the training data, and then use this training to generate new content, e.g., using transformers or other coder/decoders, that is similar to the training data; large language models have also been used in chatbots and virtual assistants , where they can understand and respond to human language input in a natural and conversational way; these models can encode and benefit from broad knowledge bases and expertise in various domains on which the model was trained, and may provide a natural language or other intuitive interface for user interactions; ¶¶ [0077]-[0079]: generative AI may be used with user-provided data and a large language model to provide summarization, outlining, information extraction, information expansion (e.g., prompted content creation) , recommendations, translation, rephrasing, sentiment analysis, coding, text to programmatic commands, code to natural language, natural language to database query (e.g., SQL), classification, grammar correction, conversational interaction (e.g., for chat bots or the like), and so forth; adapt inferencing based on machine learning for use in a manufacturing context; existing pre-trained models may be used sequentially or iteratively to prepare manufacturing data for use with existing foundation models, e.g. by initially requesting a data summary , and then presenting this summary along with a specific request for analysis or recommendations; ¶¶ [0083]-[0092] with 302 and 304 in FIG. 3: converting a range of data for a manufacturing process into a form that can be ingested by a generative AI engine, and providing suitable prompts to the generative AI engine to summarize, analyze, and make recommendations based on the data; in step 302, receiving data such as manufacturing data from a manufacturing system or other process, facility or the like; in step 304, converting manufacturing data into an intermediate representation; e.g., converting the manufacturing data into a natural language representation or near natural language representation including a human-readable description of the manufacturing data, where a natural language representation would include language closely mimicking human language, a near natural language representation permits the use of more encoded data in text form, e.g., using schemas, fields, or the like; e.g., structured queries, Boolean operators, wildcards, and other logical operators, program command syntaxes, and the like, may be incorporated into a text-based near natural language representation to provide greater flexibility in the format of data inputs; implicit metadata may include data such as an inferred type for the application, a programming context for the application, and so forth; the type may be inferred using the classification techniques; permit the type of an application, as well as other context, to be inferred based on, e.g., connections to other applications, the type of data received, the type of commands issued, the structure and content of a user interface for the application, a static or behavioral analysis of the application, and so forth; the manufacturing data may also or instead include process data from at least one sensor in the manufacturing process, such as a sensor controlled by the application; this may include sensor data such as discrete or time series measurements of, e.g., temperature, weight, speed, force, voltage, strain, and so forth; sensor data may be raw sensor data, filtered sensor data, processed sensor data (e.g., via descriptive statistics or the like), and so forth; the intermediate representation may be any natural language characterization or near natural language characterization that allows an inference engine to use a pre-trained large language model (such as ChatGPT 3, ChatGPT 3.5, ChatGPT 4, Google's Bidirectional Encoder Representations from Transformers (BERT), and so forth) or the like to process or reason about the data semantically, bringing in knowledge from across a wide domain of training materials; the inference engine may also be finetuned on data from a specific domain , such as a particular industry, academic or scientific discipline, and so forth; ¶¶ [0093]-[0094] with 306 in FIG. 3: in step 306, requesting a summarization of the (near) natural language representation from a first model, such as by requesting a summarization of a human-readable description from a first large language model; the first model may more generally include any suitable foundation model, such as a language model or a large language model pre-trained on a large corpus of text; the model may also or instead include a refined large language model, or a language model trained on, e.g., domain-specific content for a manufacturing environment or the like; the model may be also or instead include a student model trained with any of the above, or any other compressed model or the like suitable for, e.g., edge computing, event processing, or other deployment in a local context; facilitating an automated conversion of the intermediate representation into a natural language description using the processing power and knowledge encoded in the large language model; in order to request a summarization, the intermediate representation of the manufacturing data may be presented to the model, and the model may be requested, e.g., with a suitable prompt or the like, to generate a description or summarization of the intermediate representation; the request for a summarization may also be further parameterized using natural language or near natural language to manage a prompt to a language model or the like, e.g., to request specific analysis or recommendations, or to focus the analysis on particular areas of interest; e.g., the input to the model may request initial insights relating to, e.g., possible data anomalies or result effective variables , or may provide instructions about how to organize or summarize the data; ¶¶ [0095]-[0098] with 308 in FIG. 3: in step 308, requesting an analysis of the summarization; presenting the summarization from the first model to a second model, along with a request for an analysis; the request for analysis may be presented as a prompt to a language model, or as any other suitable request formatted for the model as appropriate; the request may include a request for one or more result effective parameters for the manufacturing process, e.g., variables that affect the speed, cost, quality, or other characteristic(s) of the process; the request may include a request for possible improvements, quality control measures, modifications, and so forth; the request for analysis may request specific statistical analyses (mean, standard deviation, etc.), comparisons (e.g., compare operator 1 to operator 2, or manufacturing line 1 to manufacturing line 2), optimizations (e.g., "How do I make this faster?" or "How do I reduce error rates?"), or other analysis of result effective variables and the like; a user can also ask for recommendations to improve the process, or for recommendations on suitable database queries, code revisions, and so forth; these latter recommendations may be parsed back to SQL, user interface reconfigurations, application code, or the like by the model, or may be presented in text form for use by a user that receives a corresponding result; requesting a summarization (of near natural language description) from a first model (such as a first language model); one of the models may be a different model, e.g., a language model trained or fine-tuned with domain-specific training data for the manufacturing environment or some other knowledge domain ; e.g., by refining the first model for the current manufacturing environment, a summarization can be rendered that more accurately expresses the operating environment for downstream use by non-domain-specific inference engines; the request may include a request for a pie chart, graph, time series comparison, or other type of visual output based on the data input and any intervening analysis by the large language model; ¶¶ [0102]-[0107] with FIG. 4: the manufacturing environment 402 may generate manufacturing data from, e.g., sensors, human inputs, applications, and the like that are associated with a manufacturing process as a data stream or as raw data; this data may be filtered, augmented, aggregated, or otherwise processed for storage and/or for subsequent use in generating recommendations; manufacturing data may also or instead include external data such as process documentation, user manuals, third party data, customer requirements, product specifications, regulations, industry standards, inventory and price data, process images, process videos, and so forth; the implicit metadata may include, e.g., inferred data about the application that is not explicitly encoded in the source(s ), but can be inferred based on other available information; this may be converted into text form for use in natural language processing, e.g., by converting numbers to text, or providing descriptors for data tables, time series data, and so forth; sensor metadata may include context for a sensor that describes, e.g., intrinsic properties of the sensor (units of measurement, range of measurement, accuracy, sampling rates, etc.) or information about how the sensor fits into a manufacturing process (e.g., location, purpose, connected applications, and so forth); the data converter 404 may receive the manufacturing data from the manufacturing environment 402 and generate a near natural language description of the manufacturing data for use by a language model; the data converter 404 may apply scaling strategies to manage the quantity and quality of data that is presented to inference engines, which may include filtering real time data to a time scale of interest, and labeling data according to a type of event, source of event, and so forth, in order to compress the data that requires processing; the data converter 404 may extract statistical descriptions (minimum, maximum, range, mean , median , mode , variance , trend analysis, regression parameters , etc.); the data converter 404 may hierarchically or iteratively apply a language model, e.g., to summarize groups of events in a related meta-segment of data, and then to summarize a group of such summaries, in order to distill the raw data before initiating an analysis; a first configuration module 406 may be used to present the converted data from the manufacturing environment 402 to a first inference engine 408, such as a language model, large language model, refined language model, student model, or the like; a language model such as a large language model may be used to advantageously support inferences based on a knowledge domain derived from a large corpus of text-based documents upon which it was trained; the first configuration module 406 may be configured to request a natural language summary of the data from the manufacturing environment 402, e.g., by presenting the near natural language description from the data converter 404 to a language model or other first inference engine 408 along with suitable prompts; ¶ [0112] with FIG. 4: The user request management module 418 may provide a user-friendly interface for creating requests on one hand, while managing the first configuration module 406 and the second configuration module 410 so that suitable data and prompts (or the like) are presented to the inference engines to generate a response; e.g., the user interface 416 may present available data sources from the manufacturing environment 402 as check boxes, drop down lists, or the like so that a user can select what data is to be used when performing an analysis; the user request management module 418 may receive the various parameters from a user, and may use these parameters to configure the first configuration module 406 and the second configuration module 410 to generate suitable, corresponding prompts for use by the inference engines in generating a response to the request received from the user; ¶¶ [0116]-[0126] with FIG. 5: in step 502, identifying an application, such as an application that controls a manufacturing process; in step 506, segmenting the application into steps for purposes of characterizing an execution time for the application; programmatically inferring process steps and creating the execution graph based on an analysis of one or more of the software modules in an application; programmatically inferring a step based on user inputs to the software module, or based on features or functions of a user interface for the application; e.g., where user interaction is requested by a widget or the like, e.g., to initiate an operation, to indicate approval of a result, to request an inspection, and so forth, these programmatic demarcations may be identified and associated with steps in the process; create an embedding, feature vectors, parameters, or the like for evaluating similarity to other steps to facilitate analysis of a new, unknown application; other context (e.g., triggers for actions within the application, variables maintained and updated by the application, records captured by the application, machines and/or devices controlled by or monitored by the application, connectors to other applications and process entities, and so forth) may also be used to infer segmentation and identify process steps; each interaction point in an execution graph or other data structure describing these elements provides a potential demarcation for a process step that may be modeled for step timing; the steps in the application, or in the process controlled by the application, may include one or more steps controlled directly by the application, e.g., where the application autonomously executes a process step based on timing, sensor feedback, trigger events, and so forth; the steps may also or instead include one or more steps controlled by a user of the application, e.g., where the user interface prompts a user for input during the process; a preliminary time estimate for each step can also help to converge more quickly and efficiently at an accurate characterization of the process or help to identify errors in segmentation; an initial timing expectation for an application, or for one of the steps, may be obtained from a variety of sources; the timing expectation may be based on manual user steps in the process, prior observations, design expectations , regulations or industry standards applicable to the process, and so forth; in step 508, monitoring execution of the application; applications for a manufacturing process will run in real time, and data may be collected in substantially real time for any monitored aspects of the process; the realized timings for each step may be collected over any suitable window of time, and at any suitable temporal resolution, and may be based on any monitored inputs such as sensor feedback, user input to an application prompt , machine vision, and so forth; thus, monitoring may include acquiring timing data such as a timing data distribution for each step, and/or one or more other statistical descriptors or other quantitative descriptions; in step 510, determining whether there are additional applications to be analyzed; if there are additional applications, return to step 502, where the next application is identified and processed (e.g., for segmentation and monitoring); if there are no additional applications, proceed to step 512 where an execution model can be created; ¶ [0129]: the user interface (represented as screen shots, images, user interface code, or the like) may provide a basis for identifying steps and/or creating an embedding in a manner that facilitates step time estimation for a new process; the regression parameters for the regression model may also be selected based on the embedding, e.g., by selecting regression parameters based on similarity to other applications (or other process descriptions); ¶ [0133]: permits the creation of company-specific, industry-specific, or process-specific databases of step time models for expected execution timing; certain user interface elements, widgets, or functions may be associated with particular execution models, and corresponding step timings; similarly, a particular type of step (e.g., insert four screws with a T9 Torx™ screwdriver) may have similar timing expectations independent of the workpiece or manufacturing context; certain types of processes have steps prescribed by regulation, industry standard, or common practice, and may usefully be modeled across companies and/or processes; e.g., a safety audit, pharmaceutical line clearance, or visual part inspection may have similar or identical use cases across users; different application categories may be used to estimate timing; e.g., an application may be categorized as "assembly" (simple, moderate, complex, ... ), machining, quality control, etc., and this may be used to establish an initial timing estimate for a particular application, or to parameterize process steps within a latent space or embedding, which may also include mixed-type categories (e.g., 80% assembly, 20% quality control), each of which may receive a separate execution model, or separate embeddings when evaluating an unknown application for purposes of estimating execution timing; these characteristics may be determined based on descriptive metadata, inferred from characteristics of an application, or using any of the other techniques; ¶ [0139]: the user interface may facilitate use of the execution model 522 by permitting a user to provide or identify a new application , and to apply the execution model 522 to the new application for derivation of an execution time estimate ; the user interface may also facilitate subsequent steps such as estimate validation , estimate refinement , and so forth, as well as preceding steps such as creation and tuning of the execution model 522; ¶ [0156]: receiving a new application, such as an unknown application that has yet to be classified). Claims 5, 15, and 20 Linder discloses all the elements as stated in Claim 1 and further discloses determining a preferred output format for the user based on user data associated with the user, wherein constructing the second prompt comprises appending the user data to the one or more prompts, the one or more responses, and the instruction string, and the instruction string comprises instructions to the generative model to determine a preferred output format for the user based on the user data and to present the interpretation in the preferred output format as the first content (Linder, ¶ [0098]: the request may include particular formats for summary data; e.g., the request may include a request for a pie chart, graph, time series comparison , or other type of visual output based on the data input and any intervening analysis by the large language model; ¶ [0077]: an inference engine based on generative AI can be used to create recommendations by analyzing patterns and similarities in user-provided datasets of behaviors or preferences , and then generating new recommendations based on that analysis; e.g., generative AI may be used with user-provided data and a large language model to provide summarization, outlining, information extraction, information expansion (e.g., prompted content creation), recommendations, translation, rephrasing, sentiment analysis, coding, text to programmatic commands, code to natural language, natural language to database query (e.g., SQL), classification, grammar correction, conversational interaction (e.g., for chat bots or the like), and so forth; generative AI can be used to analyze user behavior data, instrumentation or other process data, productivity and results, and to generate new recommendations, tools, analysis, and the like based in response to input data and any accompanying prompts , requests , analysis parameters , or the like; ¶¶ [0080]-[0082]: support a multi-model head end for a generative AI system; by fusing these heterogenous data types into a unified representation, a suitable latent space can then be configured for creating a decoder or generator network that can create new application code based on multi-modal process descriptions; a variety of prompt engineering and other augmentation techniques may be used to improve results from a foundation model that was not trained with a domain specific data set; e.g., retrieving current (i.e., non-training) data or other descriptive text and the like from a manufacturing process to augment a request from an inference engine, and/or augmenting model outputs by retrieving process-specific data relevant to one or more portions of the generative output; constrain a generative AI analysis or coding task to a particular manufacturing domain or process in order to facilitate the creation of useful results without requiring a full custom foundation model for, e.g., coding manufacturing applications; ¶¶ [0092]-[0094] with FIG. 3: while the method 300 shown in FIG. 3 focuses on text-based natural language processing, similar generative AI models may be created for other modes of input and/or output such as audio, speech, video, images, and so forth; the intermediate representation may usefully include any of the foregoing, provided the corresponding generative AI system can interpret the data; extracting domain-specific features of interest to augment a prompt or other request to an inference engine that uses a more generic foundation model; supplemental information may be explicitly provided by a user, or may be automatically generated using a domain-specific embedding, retrieval-augmented generation (RAG), or other techniques to automatically generate (or locate) and append supplemental information to the intermediate representation; ¶¶ [0096]-[0098]: requesting the analysis may include requesting computer readable instructions implementing one or more recommendations contained in the analysis; e.g., the input to the generative AI model may include a request to identify the operator with the lowest error rate, and recommend code changes in an application based on the behavior of that operator; many current large language models can generate executable code based on task descriptions or other specifications, and can be specifically requested to produce code, e.g., in a particular programming language or for a particular interpreter or environment; more refined models may also or instead be used to generate code suitable for a particular manufacturing context based on, e.g., available libraries, connectors, resources, code bases, and so forth; ¶ [0101] with 312 in FIG. 3: in step 312, performing further processing on the analysis; post-processing the output for presentation, e.g., for normalization, for consistency with terminology used in the manufacturing process, to format or organize the output data, to augment the results, and so forth; presenting recommended coding changes, along with an explanation for the nature of, and reasons for, the recommended changes; ¶ [0104] with FIG. 4: descriptive metadata may be used to create natural language characterizations of applications, application data, application logs, sensors, sensor data, and so forth, so that the data can analyzed within the context of the manufacturing process, which permits an inference engine such as a large language model to analyze and interpret manufacturing data in the correct context , rather than simply as a collection of structured or unstructured quantitative data; in particular, the metadata for the sensor(s) can impart physical meaning to quantitative data , and metadata for the application(s) can permit inferences about, or reasoning based on, the intended use and nature of the sensor data; a variety of techniques for automated data augmentation may be used to augment raw sensor/application outputs with suitable metadata to assist in interpretation for the purposes described herein; ¶¶ [0110]-[0111] with FIG. 4: a presentation module 414 may receive the recommendation from the second inference engine 412, and may perform any supplemental post-processing useful for presenting the results to a user; e.g., the presentation module 414 may receive a natural language recommendation from a second language model, and may parse the recommendation for presentation in a user interface 416 according to one or more user criteria ; this may include filtering the recommendation, formatting the recommendation for presentation in a user interface 416 (e.g., as one or more windows showing relevant data, analysis, recommendations, code implementing recommendations, and so forth); e.g., language in the recommendation may be mapped to vocabulary, process descriptions, and the like from process documentation so that the results are cast in a rubric that can be readily understood and interpreted by a human reviewer who is familiar with the manufacturing environment 402; the presentation module 414 may usefully manage queries to the data store 403, e.g., to access raw data from the manufacturing environment 402 to augment analysis and recommendations that are output by the second inference engine 412, which permits inference based on a reduced-size, abstracted data set, while permitting augmentation with more granular data once a particular recommendation is identified; ¶ [0130]: matrix factorization may then be performed to identify the latent factors that explain variation in execution times across the different types of steps; these latent factors may then be used to predict a total execution time for an application (or other unit of process) based on constituent elements; ¶¶ [0151]-[0152]: text description of an application may be augmented with supplemental information to support multi-modal classification; ¶¶ [0157]-[0162] with FIG. 7: converting the new application into a text-based format or other natural language description; receiving a supplemental description of the new application which include any of the supplemental or augmented descriptions or data matched to the type of supplemental data used to train the classification engine 722; generating an embedding for the supplemental data to extract features, and/or transform the supplemental description for the new application into a vector or set of vectors that can be used as inputs for classification; classifying the new application by applying the classification engine to the new application, e.g., based on the description and the supplemental description, and/or any associated embeddings; the output of the classification engine 722 may be an application type, such as any type or category useful for managing applications in a manufacturing environment or the like; the type may also or instead be an industry category; the categories may include any category or group of functional categories, user categories, process categories, industry categories, and so forth useful for organizing and managing software in a manufacturing environment; produce an accuracy estimate such as a probability of belonging to one or more categories; a classification may also be used for any of the other application analysis or management tools, e.g., where a classification is used as supplemental information for input to another generative AI process; ¶ [0189]: in order to create a validation plan, an application may be transformed from an internal, executable representation into a text description, e.g., so that it can be compared to validation requirements; this text representation may be augmented with classification information, which may be derived from application metadata or other descriptive information, or derived by classifying the application based on the content and context of the application; the textualization may also or instead be augmented with images, video, and other multimedia or non-text content contained in the application, or with natural language descriptions of any of this content obtained, e.g., from suitable embeddings and/or inference engines; ¶ [0193]: permits the description to be augmented with additional data about the type or context of the application that are inferred from characteristics of the application, e.g., so that appropriate, corresponding testing or validation requirements can be identified and applied; ¶ [0195]: a wide range of supplemental data may be provided and used to augment application descriptions on one hand, and create an embedding used to characterize applications on the other; supplemental data may be acquired from a manufacturing environment such as individual user tendencies observed during monitoring, or user characteristics such as age, experience, efficiency, physical size, and so forth; the supplemental data may include physical characteristics of a workspace, such as a working volume, dimensions of a desk, locations of bins, screens, input/output devices, and so forth; these extra modalities, which may be provided as physical specifications, images, text descriptions, and the like, may be used to create a multimodal representation of an application or process that extends well beyond code and machinery, and can improve an embedding space used to extract relevant features from existing testing/validation requirements and to create testing or validation plans; ¶ [0200]: where new code or coding modifications are included in the test plan, these tools may be used, e.g., by applying natural language processing techniques or the like to interpret a description of a process, applying inference engines such as large language models to identify key components of the description (such as variables, functions, conditions, loops, etc.) that need to be converted into code, and then mapping these components to programming constructs). Claim 6 Linder discloses all the elements as stated in Claim 1 and further discloses generating summaries of data in the one or more expert knowledge sources by inputting the data to the generative model, the summaries including output evaluation criteria metrics definitions, technical specifications, data on how to analyse outputs of the calculation tool, or a combination thereof associated with the statistical test, wherein executing the query on the one or more expert knowledge sources includes executing the query on the summaries (Linder, ¶ [0069] with FIG. 2: the content sources 210 may include any sources of content for use herein; this may, e.g., include industry data, manufacturing specifications and standards, user guides, code repositories, device specifications, customer requirements, databases, collaborative resources, inventory or price data, and so forth; ABSTRACT and ¶ [0083]: data from manufacturing is highly uncontextualized and siloed, requiring expert knowledge of context and substantial data pre-processing to support meaningful queries and visualizations; data for a number of different sources in a manufacturing context/environment can be retrieved and converted into an intermediate representation in a natural language or near-natural language form, which can in tum be ingested by a generative AI engine, along with suitable prompts by the user to summarize , analyze, and make recommendations based on the data; ¶¶ [0073]-[0074]: generative AI technologies use machine learning models such as deep learning networks to analyze vast amounts of data, based upon which they can learn patterns, styles, or rules, or otherwise encode knowledge contained in the training data, and then use this training to generate new content, e.g., using transformers or other coder/decoders, that is similar to the training data; large language models have also been used in chatbots and virtual assistants , where they can understand and respond to human language input in a natural and conversational way; these models can encode and benefit from broad knowledge bases and expertise in various domains on which the model was trained, and may provide a natural language or other intuitive interface for user interactions; ¶¶ [0077]-[0079]: generative AI may be used with user-provided data and a large language model to provide summarization , outlining , information extraction , information expansion (e.g., prompted content creation) , recommendations, translation, rephrasing, sentiment analysis, coding, text to programmatic commands, code to natural language, natural language to database query (e.g., SQL), classification , grammar correction, conversational interaction (e.g., for chat bots or the like), and so forth; generative AI can employ content-based filtering, which involves analyzing the characteristics of items to make recommendations that a user may find helpful; generative AI can be used to analyze manufacturing data , user actions , and so forth to generate new recommendations for operating a manufacturing line or the like based on similarities to other manufacturing processes ; a hybrid approach may also be used, where recommendations combine collaborative filtering and content-based filtering to make recommendations; generative AI can be used to analyze user behavior data , instrumentation or other process data , productivity and results , and to generate new recommendations , tools , analysis , and the like based in response to input data and any accompanying prompts, requests, analysis parameters, or the like; adapt inferencing based on machine learning for use in a manufacturing context ; existing pre-trained models may be used sequentially or iteratively to prepare manufacturing data for use with existing foundation models, e.g. by initially requesting a data summary , and then presenting this summary along with a specific request for analysis or recommendations ; ¶¶ [0083]-[0092] with 302 and 304 in FIG. 3: converting a range of data for a manufacturing process into a form that can be ingested by a generative AI engine, and providing suitable prompts to the generative AI engine to summarize , analyze, and make recommendations based on the data; in step 302, receiving data such as manufacturing data from a manufacturing system or other process, facility or the like; in step 304, converting manufacturing data into an intermediate representation; e.g., converting the manufacturing data into a natural language representation or near natural language representation including a human-readable description of the manufacturing data, where a natural language representation would include language closely mimicking human language, a near natural language representation permits the use of more encoded data in text form, e.g., using schemas, fields, or the like; e.g., structured queries, Boolean operators, wildcards, and other logical operators, program command syntaxes, and the like, may be incorporated into a text-based near natural language representation to provide greater flexibility in the format of data inputs; implicit metadata may include data such as an inferred type for the application, a programming context for the application, and so forth; the type may be inferred using the classification techniques; permit the type of an application, as well as other context, to be inferred based on, e.g., connections to other applications, the type of data received, the type of commands issued, the structure and content of a user interface for the application, a static or behavioral analysis of the application, and so forth; the manufacturing data may also or instead include process data from at least one sensor in the manufacturing process, such as a sensor controlled by the application; this may include sensor data such as discrete or time series measurements of, e.g., temperature, weight, speed, force, voltage, strain, and so forth; sensor data may be raw sensor data, filtered sensor data, processed sensor data (e.g., via descriptive statistics or the like), and so forth; the intermediate representation may be a narrative summary of a data table retrieved in response to a structured query; the intermediate representation may be any natural language characterization or near natural language characterization that allows an inference engine to use a pre-trained large language model (such as ChatGPT 3, ChatGPT 3.5, ChatGPT 4, Google's Bidirectional Encoder Representations from Transformers (BERT), and so forth) or the like to process or reason about the data semantically, bringing in knowledge from across a wide domain of training materials; the inference engine may also be finetuned on data from a specific domain , such as a particular industry, academic or scientific discipline, and so forth; ¶¶ [0093]-[0094] with 306 in FIG. 3: in step 306, requesting a summarization of the (near) natural language representation from a first model, such as by requesting a summarization of a human-readable description from a first large language model; the first model may more generally include any suitable foundation model, such as a language model or a large language model pre-trained on a large corpus of text; the model may also or instead include a refined large language model, or a language model trained on, e.g., domain-specific content for a manufacturing environment or the like; the model may be also or instead include a student model trained with any of the above, or any other compressed model or the like suitable for, e.g., edge computing, event processing, or other deployment in a local context; facilitating an automated conversion of the intermediate representation into a natural language description using the processing power and knowledge encoded in the large language model; in order to request a summarization , the intermediate representation of the manufacturing data may be presented to the model, and the model may be requested, e.g., with a suitable prompt or the like, to generate a description or summarization of the intermediate representation; the request for a summarization may also be further parameterized using natural language or near natural language to manage a prompt to a language model or the like, e.g., to request specific analysis or recommendations, or to focus the analysis on particular areas of interest; e.g., the input to the model may request initial insights relating to, e.g., possible data anomalies or result effective variables , or may provide instructions about how to organize or summarize the data; ¶¶ [0095]-[0098] with 308 in FIG. 3: in step 308, requesting an analysis of the summarization ; presenting the summarization from the first model to a second model, along with a request for an analysis; the request for analysis may be presented as a prompt to a language model, or as any other suitable request formatted for the model as appropriate; the request may include a request for one or more result effective parameters for the manufacturing process, e.g., variables that affect the speed , cost , quality , or other characteristic(s) of the process ; the request may include a request for possible improvements, quality control measures, modifications, and so forth; the request for analysis may request specific statistical analyses (mean, standard deviation, etc.), comparisons (e.g., compare operator 1 to operator 2, or manufacturing line 1 to manufacturing line 2), optimizations (e.g., "How do I make this faster?" or "How do I reduce error rates?"), or other analysis of result effective variables and the like; after obtaining a summarization in natural language form from the generative AI system, a second model can be used to analyze or reason about the summarization, or about multiple summarizations concurrently; e.g., given a stack of summarizations of an execution of an application ("Alice completed app A in an average of 30 seconds with an average of 5 defects per day , Bob completed app A in an average of 36 seconds with an average of 3 defects per day , ... "), a request to the model for analysis may be stated as "Who is the best operator on app A in terms of time?" or "Who is the best operator in terms of defects?"; a user can also ask for recommendations to improve the process, or for recommendations on suitable database queries, code revisions, and so forth; these latter recommendations may be parsed back to SQL, user interface reconfigurations, application code, or the like by the model, or may be presented in text form for use by a user that receives a corresponding result; requesting a summarization (of near natural language description) from a first model (such as a first language model); one of the models may be a different model, e.g., a language model trained or fine-tuned with domain-specific training data for the manufacturing environment or some other knowledge domain ; e.g., by refining the first model for the current manufacturing environment, a summarization can be rendered that more accurately expresses the operating environment for downstream use by non-domain-specific inference engines; the input to the generative AI model may include a request to identify the operator with the lowest error rate , and recommend code changes in an application based on the behavior of that operator; the request may include a request for a pie chart, graph, time series comparison, or other type of visual output based on the data input and any intervening analysis by the large language model; ¶¶ [0102]-[0107] with FIG. 4: the manufacturing environment 402 may generate manufacturing data from, e.g., sensors, human inputs, applications, and the like that are associated with a manufacturing process as a data stream or as raw data; this data may be filtered, augmented, aggregated, or otherwise processed for storage and/or for subsequent use in generating recommendations; manufacturing data may also or instead include external data such as process documentation, user manuals, third party data, customer requirements, product specifications, regulations, industry standards, inventory and price data, process images, process videos, and so forth; the implicit metadata may include, e.g., inferred data about the application that is not explicitly encoded in the source(s ), but can be inferred based on other available information; this may be converted into text form for use in natural language processing, e.g., by converting numbers to text, or providing descriptors for data tables, time series data, and so forth; sensor metadata may include context for a sensor that describes, e.g., intrinsic properties of the sensor (units of measurement, range of measurement, accuracy, sampling rates, etc.) or information about how the sensor fits into a manufacturing process (e.g., location, purpose, connected applications, and so forth); the data converter 404 may receive the manufacturing data from the manufacturing environment 402 and generate a near natural language description of the manufacturing data for use by a language model; the data converter 404 may apply scaling strategies to manage the quantity and quality of data that is presented to inference engines, which may include filtering real time data to a time scale of interest, and labeling data according to a type of event, source of event, and so forth, in order to compress the data that requires processing; the data converter 404 may extract statistical descriptions (minimum, maximum, range, mean, median, mode, variance, trend analysis, regression parameters, etc.); the data converter 404 may hierarchically or iteratively apply a language model, e.g., to summarize groups of events in a related meta-segment of data, and then to summarize a group of such summaries , in order to distill the raw data before initiating an analysis; a first configuration module 406 may be used to present the converted data from the manufacturing environment 402 to a first inference engine 408, such as a language model, large language model, refined language model, student model, or the like; a language model such as a large language model may be used to advantageously support inferences based on a knowledge domain derived from a large corpus of text-based documents upon which it was trained; the first configuration module 406 may be configured to request a natural language summar y of the data from the manufacturing environment 402, e.g., by presenting the near natural language description from the data converter 404 to a language model or other first inference engine 408 along with suitable prompts; ¶ [0112] with FIG. 4: The user request management module 418 may provide a user-friendly interface for creating requests on one hand, while managing the first configuration module 406 and the second configuration module 410 so that suitable data and prompts (or the like) are presented to the inference engines to generate a response; e.g., the user interface 416 may present available data sources from the manufacturing environment 402 as check boxes, drop down lists, or the like so that a user can select what data is to be used when performing an analysis; the user request management module 418 may receive the various parameters from a user, and may use these parameters to configure the first configuration module 406 and the second configuration module 410 to generate suitable, corresponding prompts for use by the inference engines in generating a response to the request received from the user; ¶¶ [0116]-[0126] with FIG. 5: in step 502, identifying an application, such as an application that controls a manufacturing process; in step 506, segmenting the application into steps for purposes of characterizing an execution time for the application; programmatically inferring process steps and creating the execution graph based on an analysis of one or more of the software modules in an application; programmatically inferring a step based on user inputs to the software module, or based on features or functions of a user interface for the application; e.g., where user interaction is requested by a widget or the like, e.g., to initiate an operation, to indicate approval of a result, to request an inspection, and so forth, these programmatic demarcations may be identified and associated with steps in the process; create an embedding, feature vectors, parameters, or the like for evaluating similarity to other steps to facilitate analysis of a new, unknown application; other context (e.g., triggers for actions within the application, variables maintained and updated by the application, records captured by the application, machines and/or devices controlled by or monitored by the application, connectors to other applications and process entities, and so forth) may also be used to infer segmentation and identify process steps; each interaction point in an execution graph or other data structure describing these elements provides a potential demarcation for a process step that may be modeled for step timing; the steps in the application, or in the process controlled by the application, may include one or more steps controlled directly by the application, e.g., where the application autonomously executes a process step based on timing, sensor feedback, trigger events, and so forth; the steps may also or instead include one or more steps controlled by a user of the application, e.g., where the user interface prompts a user for input during the process; a preliminary time estimate for each step can also help to converge more quickly and efficiently at an accurate characterization of the process or help to identify errors in segmentation; an initial timing expectation for an application, or for one of the steps, may be obtained from a variety of sources; the timing expectation may be based on manual user steps in the process, prior observations, design expectations , regulations or industry standards applicable to the process, and so forth; in step 508, monitoring execution of the application; applications for a manufacturing process will run in real time, and data may be collected in substantially real time for any monitored aspects of the process; the realized timings for each step may be collected over any suitable window of time, and at any suitable temporal resolution, and may be based on any monitored inputs such as sensor feedback, user input to an application prompt, machine vision, and so forth; thus, monitoring may include acquiring timing data such as a timing data distribution for each step, and/or one or more other statistical descriptors or other quantitative descriptions; in step 510, determining whether there are additional applications to be analyzed; if there are additional applications, return to step 502, where the next application is identified and processed (e.g., for segmentation and monitoring); if there are no additional applications, proceed to step 512 where an execution model can be created; ¶ [0129]: the user interface (represented as screen shots, images, user interface code, or the like) may provide a basis for identifying steps and/or creating an embedding in a manner that facilitates step time estimation for a new process; the regression parameters for the regression model may also be selected based on the embedding, e.g., by selecting regression parameters based on similarity to other applications (or other process descriptions); ¶ [0133]: permits the creation of company-specific, industry-specific, or process-specific databases of step time models for expected execution timing; certain user interface elements, widgets, or functions may be associated with particular execution models, and corresponding step timings; similarly, a particular type of step (e.g., insert four screws with a T9 Torx™ screwdriver) may have similar timing expectations independent of the workpiece or manufacturing context; certain types of processes have steps prescribed by regulation, industry standard, or common practice, and may usefully be modeled across companies and/or processes; e.g., a safety audit, pharmaceutical line clearance, or visual part inspection may have similar or identical use cases across users; different application categories may be used to estimate timing; e.g., an application may be categorized as "assembly" (simple, moderate, complex, ... ), machining, quality control, etc., and this may be used to establish an initial timing estimate for a particular application, or to parameterize process steps within a latent space or embedding, which may also include mixed-type categories (e.g., 80% assembly, 20% quality control), each of which may receive a separate execution model, or separate embeddings when evaluating an unknown application for purposes of estimating execution timing; these characteristics may be determined based on descriptive metadata, inferred from characteristics of an application, or using any of the other techniques; ¶ [0138]: in general, statistical tuning may be performed over any suitable interval; the execution time estimate may be periodically compared to actual execution timing, and an update or other review may be automatically recommended when the actual execution timing deviates in some manner (e.g., beyond a threshold for maximum excursion in individual times, maximum change in a mean or median, change in the standard deviation, and so forth) from the execution time estimated based on the model). Claim 7 Linder discloses all the elements as stated in Claim 1 and further discloses inferring, via the generative model, one or more actions to take to cope with the interpretation; and causing the user interface to present the interpretation with the one or more actions (Linder, ¶ [0077]: generative AI can be used to analyze manufacturing data, user actions, and so forth to generate new recommendations for operating a manufacturing line or the like based on similarities to other manufacturing processes; ¶¶ [0096]-[0098]: a user can also ask for recommendations to improve the process , or for recommendations on suitable database queries, code revisions, and so forth; these latter recommendations may be parsed back to SQL, user interface reconfigurations, application code, or the like by the model, or may be presented in text form for use by a user that receives a corresponding result; the second model may include a large language model that is asked to generate one or more suitable SQL queries to implement the recommendations , or to generate code to implement the revisions ; requesting the analysis may include requesting computer readable instructions implementing one or more recommendations contained in the analysis; e.g., the input to the generative AI model may include a request to identify the operator with the lowest error rate, and recommend code changes in an application based on the behavior of that operator; many current large language models can generate executable code based on task descriptions or other specifications, and can be specifically requested to produce code, e.g., in a particular programming language or for a particular interpreter or environment; more refined models may also or instead be used to generate code suitable for a particular manufacturing context based on, e.g., available libraries, connectors, resources, code bases, and so forth; ¶ [0101] with 312 in FIG. 3: in step 312, performing further processing on the analysis; post-processing the output for presentation, e.g., for normalization, for consistency with terminology used in the manufacturing process, to format or organize the output data, to augment the results, and so forth; presenting recommended coding changes, along with an explanation for the nature of, and reasons for, the recommended changes ; ¶ [0104] with FIG. 4: descriptive metadata may be used to create natural language characterizations of applications, application data, application logs, sensors, sensor data, and so forth, so that the data can analyzed within the context of the manufacturing process, which permits an inference engine such as a large language model to analyze and interpret manufacturing data in the correct context , rather than simply as a collection of structured or unstructured quantitative data; in particular, the metadata for the sensor(s) can impart physical meaning to quantitative data , and metadata for the application(s) can permit inferences about, or reasoning based on, the intended use and nature of the sensor data; a variety of techniques for automated data augmentation may be used to augment raw sensor/application outputs with suitable metadata to assist in interpretation for the purposes described herein; ¶¶ [0110]-[0111] with FIG. 4: a presentation module 414 may receive the recommendation from the second inference engine 412, and may perform any supplemental post-processing useful for presenting the results to a user; e.g., the presentation module 414 may receive a natural language recommendation from a second language model, and may parse the recommendation for presentation in a user interface 416 according to one or more user criteria; this may include filtering the recommendation, formatting the recommendation for presentation in a user interface 416 (e.g., as one or more windows showing relevant data, analysis, recommendations, code implementing recommendations, and so forth); e.g., language in the recommendation may be mapped to vocabulary, process descriptions, and the like from process documentation so that the results are cast in a rubric that can be readily understood and interpreted by a human reviewer who is familiar with the manufacturing environment 402; the presentation module 414 may usefully manage queries to the data store 403, e.g., to access raw data from the manufacturing environment 402 to augment analysis and recommendations that are output by the second inference engine 412, which permits inference based on a reduced-size, abstracted data set, while permitting augmentation with more granular data once a particular recommendation is identified; ¶ [0130]: matrix factorization may then be performed to identify the latent factors that explain variation in execution times across the different types of steps; these latent factors may then be used to predict a total execution time for an application (or other unit of process) based on constituent elements). Claim 8 Linder discloses all the elements as stated in Claim 1 and further discloses wherein the user belongs to an enterprise (Linder, ¶ [0161]: displaying the classification to a user in a user interface, e.g., for review or further action, or applying a policy or management rule for an enterprise to the new application based on the classification (e.g., to submit for review, to add to a repository, to limit authorized users, etc.)), and the machine-readable storage medium further includes instructions configured to cause the processor alone or in combination with other processors to perform operations (Linder, ¶¶ [0051]-[0054] with FIG. 1: the external device 104 may be any computer or other remote resource that connects to the computing device 110 through the network 102; the external device 104 may also or instead include a network storage device, a database, a data store, a data warehouse, a cloud storage facility, a content host, or any other data storage resource or the like that the computing device 110 might usefully connect to through the network 102, e.g., for storage and retrieval of data; the computing device 110 may include a processor 112, a memory 114, a network interface 116, a data store 118, and one or more input/output interfaces 120; the processor 112 may be capable of processing instructions for execution within the computing device 110 or computer system 100; the processor 112 may be capable of processing instructions stored in the memory 114 or in the data store 118; the memory 114 may include any non-transitory computer readable medium) of: encrypting communication between the data processing system, the client device, and an enterprise system of the enterprise using a cryptographic protocol; and isolating enterprise data by keeping containers, storages, or a combination thereof at least logically or physically separate from other enterprises (Linder, ¶ [0050] wit FIG. 1: the network 102 may include public networks such as the Internet, private networks, and telecommunications networks such as the Public Switched Telephone Network or cellular networks using third generation cellular technology ( e.g., 3G or IMT-2000), fourth/fifth generation cellular technology (e.g., 4G, LTE, MT-Advanced, E-UTRA, 5G, etc.) or WiMax-Advanced (IEEE 1002 .16m)) and/or other technologies, as well as any of a variety of corporate area, metropolitan area, campus, or other local area networks or enterprise networks, along with any switches, routers, hubs, gateways, and the like that might be used to carry data among participants in the computer system 100; ¶ [0161]: applying a policy or management rule for an enterprise to the new application based on the classification to limit authorized users ; the resulting output, e.g., a type, category, or other classification for the new computer representation, may be used in any suitable manner for managing applications at a manufacturing facility or the like such as flagging an application for security review, distributing an application to appropriate personnel or data repositories , and so forth, or any of the other uses). Claim 9 Linder discloses all the elements as stated in Claim 1 and further discloses wherein the statistical test is a statistical anomaly detection test (Linder. ¶ [0094]: the input to the model may request initial insights relating to, e.g., possible data anomalies or result effective variables, or may provide instructions about how to organize or summarize the data; ¶ [0174]: anomaly thresholds may be converted into validation rules (e.g., outside diameter entered by user cannot exceed five millimeters)). Claim 10 Linder discloses all the elements as stated in Claim 1 and further discloses wherein the statistical test is associated with an incident management system (Linder, ¶ [0093]: the model may be also or instead include a student model trained with any of the above, or any other compressed model or the like suitable for, e.g., edge computing, event processing , or other deployment in a local context; ¶ [0105] with FIG. 4: filtering real time data to a time scale of interest, and labeling data according to a type of event , source of event , and so forth, in order to compress the data that requires processing; the data converter 404 may hierarchically or iteratively apply a language model, e.g., to summarize groups of events in a related meta-segment of data, and then to summarize a group of such summaries, in order to distill the raw data before initiating an analysis; other inference engines or tools may be used, e.g., to segment time series data into epochs or large-scale events , before requesting an initial summary of a group of events; ¶ [0121]: an execution graph may be any graph or similar representation or description of causal relationships, temporal relationships, or other dependencies among nodes, where each node may in turn include a state or step, e.g., of a manufacturing process or an application controlling same; e.g., each node may be a function call, instruction, data processing operation, or discrete unit of work withing the manufacturing process; the nodes may be interconnected by edges that represent transitions or flow among states ("nodes") based on an action or event such as a menu selection, button click, database or variable change, external interrupt, sensed condition, passage of time, and so forth; the edges represent dependences and indicate an order of operations between nodes; an execution graph provides a useful representation of a process, or an application to control a process, and may be explicitly provided with the application, or inferred using the techniques described herein to segment code and other data sources into an execution graph describing a process flow; ¶ [0123]: the application autonomously executes a process step based on timing, sensor feedback, trigger events, and so forth; ¶ [0171]: one of the outputs of parsing the video may include a natural language description of events in the video, which may be obtained using an AI model trained to recognize actions and provide descriptive metadata based on video input) . Conclusion 07-96 AIA The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Sharma et al. ("Statistical Data Analysis using GPT3: An Overview", 2022 IEEE Bombay Section Signature Conference (IBSSC), Dec. 8-10, 2022, pp. 1-6) discloses in Abstract of Page 1 that (1) a machine learning-based approach for statistical analysis of Data Sets; (2) proposes a novel approach for the analysis of large datasets which uses GPT3 to predict insights from calculated statistics of data; and (3) propose a novel framework to analyze large statistical data sets, which solves many computationally challenging problems in efficient ways; (3) the proposed method works on top of GPT3's features, where it learns to predict individual words from particular parts of the dataset you pass as prompts (cumulative sums/means etc.) enabling us to analyze extremely large datasets such as telecom churn or census data. Sharma further discloses in Section I of Pages 1-2 that (1) focus on how we can use GPT-3 for analyzing datasets using prompts; (2) plan to analyze three different datasets using GPT-3 and compare the results to traditional analysis methods to provide benchmarks on the accuracy of GPT-3 using a parametrical statistics approach; (3) conduct experiments on Ecommerce Sales, Heart attack, and telecom churn rate datasets; (4) for each dataset, use GPT-3 to analyze the dataset using prompts and compare its accuracy with traditional methods of analysis; (5) find that GPT-3 is able to provide reasonable insights on datasets in a very small amount of time but misses out on some information due to which it might not be as accurate as expected for every use case; (6) despite GPT-3 being trained on only language modeling objectives, these models can perform relatively well at new tasks that they need not been explicitly trained to perform, and the reason being large language models generalize to new tasks because of an implicit process of multitask learning; (7) use GPT-3 for performing analysis on a dataset using prompts; (8) decide to perform analysis on three different datasets stated in table 1 and compare the results with results from traditional analysis methods to benchmark the accuracy of analysis done by GPT-3; (9) proposed method will help analyze extremely large datasets using GPT-3 and provide insights into them; and (10) also plan to compare traditional methods for statistical analysis with the proposed method to see the pros and cons of using GPT-3 as a tool for analysis based on the performance, accuracy, and reliability of the results. Sharma also discloses in Section II in Page 2 that (1) natural language interaction features can be designed in Tableau and how it will help the user to minimize their effort while creating data visualizations using the NLI tool thereby improving UX; (2) propose GPT3 based Statistical Data Analysis Tool which takes input as dataset processed statistics than outputting the insights extracted from the data, which would otherwise require complex coding by both coders and non-coders alike saving time and efforts required for same; (3) [4] proposed an Analytics-as-a-Service (AaaS) Tool for Unstructured Data Mining; (4) the primary focus will be on using GPT3 to perform statistical analysis on structured data sets present publicly at different websites/data repositories like Kaggle or UCI Machine Learning repository available online, which can extract various features from it without much effort; (5) compared to traditional methods used earlier to analyze statistical datasets because GPT3 has shown tremendous performance when it comes to natural language processing thus making way for a new era where AI's are able to predict future outcomes with high accuracy thus making business predictions more efficient than before especially when there is no clear pattern between independent variables X1 , X; (5) [5] proposed a broader definition of big data that captures its defining characteristics, wherein their study also reinforces the need to devise new tools for predictive analytics using machine learning which is a subset of artificial intelligence in the field of computer science that often uses statistical techniques to give computers the ability to "learn" (i.e., progressively improve performance on a specific task) with data, without being explicitly programmed.; (6) the research will be based on similar grounds where analyze whether GPT can successfully solve complex statistical problems like hypothesis testing/parametric approach and make useful predictions from textual datasets or not; (7) similarly, by using datasets like Census Datal Academic Performance Scores, etc., find out whether GPT3 can be useful in solving real statistics problems or not. Moreover, our study may also help us evaluate how well does it perform compared to other existing models? Sharma further teaches in Section III of Pages 2-3 that (1) propose a practical framework for the statistical analysis of huge datasets using GPT-3 API, which requires a unique approach to prompt the zero-shot model and feed it with adequate data in order to extract useful insights from it; (2) the proposed system follows the below-mentioned steps to perform its operations: (a) Data Pre-processing; (b) Data Transformation; and (c) Analysis of Dataset; (2) GPT-3 API has the ability to extract information from raw data with the help of its zero-shot model; (3) but there is an existing limit of 16kb or 4000 tokens per request so firstly, need to pre-process the dataset in order to calculate the required statistics of the dataset; (4) these statistics are provided as prompts to GPT-3 in order to generate insights from that data; (5) as this method can be used to shrink millions or probably billions of lines of data into a statistically reasonable and easily understandable output, hence it saves us a lot of time and effort; (6) this pre-processing involves sorting the data in ascending order, taking their cumulative sum/means, median values, etc.; (7) Parameters calculation: a) Count; b) Mean; c) Median; d) Max; e) Min; f) Skewness; g) Standard Deviation; and h) Correlation Matrix; (8) the next step involves processing data into the format that is required by GPT-3, which can be achieved after performing some mathematical operations on the raw dataset; (9) we do not directly use the whole dataset for analysis but rather transform it into a more generic form like cumulative sums, means, correlations, etc.; (10) there is no specific limit of lines of data or number of columns since the GPT-3 API has zero-shot learning capabilities and doesn't require the data to be labelled for analysis, however, the total length of input for statistical data should not exceed 16kb or 4000 tokens per request; (11) also, it is imperative that pass the dataset prompts in a comma-separated format; (12) next, the dataset into the processed data is fed into GPT -3 for analysis; (13) primarily decide to process the dataset in two different ways in which GPT-3 can perform analysis on it for the best possible outcomes; (14) firstly, provide the GPT-3 API with the Correlation matrix of the dataset as one prompt and other statistical data like median, mean, standard deviation, etc. as another prompt in order to generate useful insights from it; and (15) also perform analysis on raw data by processing and labelling it before passing it into the model as prompts in order to see if the accuracy increases or not compared to processing the transformed data statistically. Sharma also teaches in Section IV of Pages 3-5 that (1) demonstrate the use of GPT-3 to automatically generate insights from statistical data; (2) specifically, show how the combination of processed statistical information (e.g., mean and standard deviation of individual columns, correlation values, etc.), attribute labels (e.g., "category" information), and prompts can be used to predict short-form phrases in the language of the author's choosing; (3) the framework used for this analysis is based on GPT-3, which enables the use of prompts to extract insights from statistical data rather than manually creating a list of insights for prediction by the model; (4) different variations of the prompts were used to influence the outputs of predicted insights, allowing for the generation of multiple sets of predictions based on statistical data; (5) overall, GPT-3 provides a sophisticated way to represent conclusions or insights as short-form phrases by taking advantage of its unique prompt/response feature; (6) GPT-3 is a language model that can generate natural outputs from inputs, which uses different types of inputs and combines them in its proposed approach, including continuous features like cumulative sums or counts along with Boolean flags indicating non-continuous information such as column data type (nominal/numeric), etc.; (7) key parameters that affect prompt outputs: (a) Temperature: it controls creativity of model. Values < 0.6 will not be creative and > 1.0 will have no context sensitivity, and hence, use values in 0.7 to 0. 9 range; (b) Top-p: controls diversity via nucleus sampling approach, and use value 1; (c) Max tokens: use top-p to prune the outputs so that only meaningful insights are generated, and use values in 400 to 500 range; (d) Frequency penalty: it controls for rare words and hence, reduces repetition by fixing the top-k value, and use 0.44; and (e) Presence penalty: it controls for repetitive words and hence, reduces repetition by fixing the top-p value in function; use 1.5, to let the model draw conclusion based on its wide domain of training data, a high presence penalty was required while tuning; (8) many different input parameters were tried out to understand what affects the performance of this research study on a sample dataset "Telecom chum rate" obtained from kaggle.com (the final experiment was carried out on real-world datasets like census and telecom chum); (9) GPT-3 is a machine learning model that takes in descriptive statistics like mean and standard deviation, as well as nominal/numeric flags, to make predictions about data; (10) in addition to these, GPT-3 can also draw conclusions based on its knowledge of whether a particular attribute value is continuous or not, along with other information such as the maximum values of each column, etc.; (11) the following are the statistical attributes used by GPT: (a) Mean: higher values imply a greater proportion of high-value records and lower mean implies more low-valued records in that column; (b) Standard deviation: standard curve distribution (bell shape) measures the variation of individual results, wherein lower values imply less deviation in results and higher values imply more deviation in results; (c) n th percentile: outliers will tell us about the presence of extremely high or low values that fall 3 times more than the standard deviation from the mean; (d) Skewness and Kurtosis: Kurtosis is used to measure how peaked a distribution is, while Skewness tells us if there are outliers in the data, wherein (i) Skewness can be used to describe the direction of skew (i.e., left-skewed vs right-skewed) in order to draw conclusions regarding the correlation between different values or deviation from mean allowed by upper bound computations like 1.5*std_dev above mean or below; (ii) Skew & Kurtosis connote information similar to that given by standard deviation but with more details on outliers; and (III) the above data briefs us about the statistical data used for insights extraction from raw data; (e) Descriptive statistics are used to summarize the characteristics of a dataset by describing or presenting features using arrays, graphs, charts, and tables of values (mean, median, etc.), wherein this approach is used to understand the dataset with minimal use of statistical procedures or methods; (12) statistical data analysis fascinates most of the existing approaches with high accuracy and reliability; (13) generate the correlation matrix and used it in prompt input; (14) there are parameters to be adjusted in GPT -3 API model; (15) in our results, decide to use code-davinci-002 and text-davinci-002, temperature=0.72, max_ tokens=527, top p= 1, frequency_penalty=0.44, presence_penalty=1.49, stop=["->Insight 11:"]; and (16) Analysis of dataset D1: a) Dataset: the dataset consists of a historical sales record of 3 branches of a supermarket company over 3 years; b) Prompt: the model was given the labeled correlation matrix and some statistics about the data and was asked to generate 10 insights on the data; c) Generated Output: (i) Product line: health and beauty have the highest mean price, so, customers may prefer to buy the most expensive products in the health and beauty category than any other product category, which may be because of the fact that people pay more attention to their body looks or even they are using costly perfumes or cosmetics which are from this product category only; (ii) Payment mode: electronic is used maximum times but its average amount is low as compared to cash payment mode, and the reason behind that can be electronic payments like debit card, credit card has lower transaction fees for processing payments thus stores provide a discount on these modes of payment which reduce the cost price and hence the net transaction value by increasing the quantity sold in case of e-payments; d) Discussions: (i) the prompt was executed 10 times and each time distinct results were generated; (ii) the generated insights were all in human-readable sentences; (iii) often the results were simply explaining a data point like two parameters being the same due to their correlation being 1; (iv) for the sales dataset, the model successfully pointed out the branches with the highest and lowest sales; (v) it was able to rank store branches according to their ratings from highest to lowest; (vi) some insights were generated on the distribution and skewness of the data; (vii) often times the generated insights were sentences explaining data points like the average custom spending and transaction amount; (viii) these insights however save the effort of visualizing the data to determine some information like understanding the most popular mode of payment; (ix) the model was able to determine the relationship between various entities of the data and successfully interpreted how a variance in the value of one entity can affect other values; (x) it also concluded that the data was not clean and had some missing values; (xi) however, this observation is incorrect because the prompt had preprocessed data which was cleaned and null values were dropped; (xi) it can be concluded that the quality of results generated depends on the quality of the data; (xii) the prompt used had two only two parameters, the correlation matrix, and the statistical data; (xiii) providing a prompt with more data points can yield in more accurate results. Any inquiry concerning this communication or earlier communications from the examiner should be directed to HWEI-MIN LU whose telephone number is (313)446-4913. The examiner can normally be reached Mon - Fri: 9:00 AM - 6:00 PM EST. Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice. If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Mariela D. Reyes can be reached at (571) 270-1006. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300. Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000. /HWEI-MIN LU/Primary Examiner, Art Unit 2142 Application/Control Number: 18/519,326 Page 2 Art Unit: 2142 Application/Control Number: 18/519,326 Page 3 Art Unit: 2142 Application/Control Number: 18/519,326 Page 4 Art Unit: 2142 Application/Control Number: 18/519,326 Page 5 Art Unit: 2142 Application/Control Number: 18/519,326 Page 6 Art Unit: 2142 Application/Control Number: 18/519,326 Page 7 Art Unit: 2142 Application/Control Number: 18/519,326 Page 8 Art Unit: 2142 Application/Control Number: 18/519,326 Page 9 Art Unit: 2142 Application/Control Number: 18/519,326 Page 10 Art Unit: 2142 Application/Control Number: 18/519,326 Page 11 Art Unit: 2142 Application/Control Number: 18/519,326 Page 12 Art Unit: 2142 Application/Control Number: 18/519,326 Page 13 Art Unit: 2142 Application/Control Number: 18/519,326 Page 14 Art Unit: 2142 Application/Control Number: 18/519,326 Page 15 Art Unit: 2142 Application/Control Number: 18/519,326 Page 16 Art Unit: 2142 Application/Control Number: 18/519,326 Page 17 Art Unit: 2142 Application/Control Number: 18/519,326 Page 18 Art Unit: 2142 Application/Control Number: 18/519,326 Page 19 Art Unit: 2142 Application/Control Number: 18/519,326 Page 20 Art Unit: 2142 Application/Control Number: 18/519,326 Page 21 Art Unit: 2142 Application/Control Number: 18/519,326 Page 22 Art Unit: 2142 Application/Control Number: 18/519,326 Page 23 Art Unit: 2142 Application/Control Number: 18/519,326 Page 24 Art Unit: 2142 Application/Control Number: 18/519,326 Page 25 Art Unit: 2142 Application/Control Number: 18/519,326 Page 26 Art Unit: 2142 Application/Control Number: 18/519,326 Page 27 Art Unit: 2142 Application/Control Number: 18/519,326 Page 28 Art Unit: 2142 Application/Control Number: 18/519,326 Page 29 Art Unit: 2142 Application/Control Number: 18/519,326 Page 30 Art Unit: 2142 Application/Control Number: 18/519,326 Page 31 Art Unit: 2142 Application/Control Number: 18/519,326 Page 32 Art Unit: 2142 Application/Control Number: 18/519,326 Page 33 Art Unit: 2142 Application/Control Number: 18/519,326 Page 34 Art Unit: 2142 Application/Control Number: 18/519,326 Page 35 Art Unit: 2142 Application/Control Number: 18/519,326 Page 36 Art Unit: 2142 Application/Control Number: 18/519,326 Page 37 Art Unit: 2142 Application/Control Number: 18/519,326 Page 38 Art Unit: 2142 Application/Control Number: 18/519,326 Page 39 Art Unit: 2142 Application/Control Number: 18/519,326 Page 40 Art Unit: 2142 Application/Control Number: 18/519,326 Page 41 Art Unit: 2142 Application/Control Number: 18/519,326 Page 42 Art Unit: 2142 Application/Control Number: 18/519,326 Page 43 Art Unit: 2142 Application/Control Number: 18/519,326 Page 44 Art Unit: 2142 Application/Control Number: 18/519,326 Page 45 Art Unit: 2142 Application/Control Number: 18/519,326 Page 46 Art Unit: 2142 Application/Control Number: 18/519,326 Page 47 Art Unit: 2142 Application/Control Number: 18/519,326 Page 48 Art Unit: 2142 Application/Control Number: 18/519,326 Page 49 Art Unit: 2142 Application/Control Number: 18/519,326 Page 50 Art Unit: 2142 Application/Control Number: 18/519,326 Page 51 Art Unit: 2142 Application/Control Number: 18/519,326 Page 52 Art Unit: 2142 Application/Control Number: 18/519,326 Page 53 Art Unit: 2142 Application/Control Number: 18/519,326 Page 54 Art Unit: 2142 Application/Control Number: 18/519,326 Page 55 Art Unit: 2142 Application/Control Number: 18/519,326 Page 56 Art Unit: 2142 Application/Control Number: 18/519,326 Page 57 Art Unit: 2142 Application/Control Number: 18/519,326 Page 58 Art Unit: 2142 Application/Control Number: 18/519,326 Page 59 Art Unit: 2142 Application/Control Number: 18/519,326 Page 60 Art Unit: 2142 Application/Control Number: 18/519,326 Page 61 Art Unit: 2142 Application/Control Number: 18/519,326 Page 62 Art Unit: 2142 Application/Control Number: 18/519,326 Page 63 Art Unit: 2142 Application/Control Number: 18/519,326 Page 64 Art Unit: 2142 Application/Control Number: 18/519,326 Page 65 Art Unit: 2142 Application/Control Number: 18/519,326 Page 66 Art Unit: 2142 Application/Control Number: 18/519,326 Page 67 Art Unit: 2142 Application/Control Number: 18/519,326 Page 68 Art Unit: 2142 Application/Control Number: 18/519,326 Page 69 Art Unit: 2142 Application/Control Number: 18/519,326 Page 70 Art Unit: 2142 Application/Control Number: 18/519,326 Page 71 Art Unit: 2142 Application/Control Number: 18/519,326 Page 72 Art Unit: 2142 Application/Control Number: 18/519,326 Page 73 Art Unit: 2142 Application/Control Number: 18/519,326 Page 74 Art Unit: 2142 Application/Control Number: 18/519,326 Page 75 Art Unit: 2142 Application/Control Number: 18/519,326 Page 76 Art Unit: 2142 Application/Control Number: 18/519,326 Page 77 Art Unit: 2142 Application/Control Number: 18/519,326 Page 78 Art Unit: 2142 Application/Control Number: 18/519,326 Page 79 Art Unit: 2142 Application/Control Number: 18/519,326 Page 80 Art Unit: 2142 Application/Control Number: 18/519,326 Page 81 Art Unit: 2142 Application/Control Number: 18/519,326 Page 82 Art Unit: 2142 Application/Control Number: 18/519,326 Page 83 Art Unit: 2142 Application/Control Number: 18/519,326 Page 84 Art Unit: 2142 Application/Control Number: 18/519,326 Page 85 Art Unit: 2142 Application/Control Number: 18/519,326 Page 86 Art Unit: 2142 Application/Control Number: 18/519,326 Page 87 Art Unit: 2142 Application/Control Number: 18/519,326 Page 88 Art Unit: 2142 Application/Control Number: 18/519,326 Page 89 Art Unit: 2142 Application/Control Number: 18/519,326 Page 90 Art Unit: 2142
Prosecution Insights