DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Response to Arguments
Applicant's arguments filed 5/13/2026 have been fully considered
35 USC § 102 & 35 USC § 103:
Regarding Applicant’s Argument (pages: 9-12): Examiner’s response:- Regarding arguments and corresponding new amendments on the independent claims 1 & 12, Applicant’s arguments with respect to the rejection(s) of under 35 USC § 102/103 have been fully considered, upon further consideration a new ground(s) of rejection is made in view of US 20250190449 A1; Zhang; Bo et al. (hereinafter Zhang). Regarding arguments and corresponding new amendments on the independent claim 13, Applicant’s arguments with respect to the rejection(s) of under 35 USC § 102/103 have been fully considered, upon further consideration a new ground(s) of rejection is made in view of US 20250190449 A1; Zhang; Bo et al. (hereinafter Zhang) & US 20150379427 A1; DIRAC; LEO PARKER et al (hereinafter Dirac). It is important to note that due to the amendments this rejection has now become a 103 obviousness rejection and is no longer a 102 anticipation rejection, henceforth there can be obviousness conclusions reached in the mapping and teaching of the prior arts inventions into the instant applications claim limitations. Newly cited prior art Zhang cites in paragraph 31 "AI agents 126 include any types of generative AI agents that are capable of performing analytics or mathematical operations on input data, such as a comma-separated value (CSV) agent, a structured query language (SQL) ..." While primary art Arriaga may lack the explicit language of SQL it clearly shows structured querying ability from figure 1 and Arriaga paragraphs 22 which states "software program where the commands to be executed are determined in advance by a user and written in a programming language. A machine-learning process may utilize supervised, unsupervised, lazy-learning processes and/or neural networks, described further below.", paragraph 24 which states "FIG. 1, apparatus 100 may include a database 116. Database 116 may be implemented, without limitation, as a relational database...in a relational database. Persons skilled in the art, upon reviewing the entirety of this disclosure, will be aware of various ways in which data entries in database may store, retrieve, organize, and/or reflect data and/or records." Figure 1 of Arriaga shows a structure for searching/querying using programs on a retrieval in a database with structure. Regarding arguments and corresponding new amendments on new claim 21, Applicant’s arguments with respect to the rejection(s) of under 35 USC § 102/103 have been fully considered, upon further consideration a new ground(s) of rejection is made in view of US 20150379427 A1; DIRAC; LEO PARKER et al (hereinafter Dirac)
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claim 1-8 and 10-12 are rejected under 35 U.S.C. 103 as being unpatentable over US 20250054068 A1; Arriaga; Jason (hereinafter Arriaga) in view of US 20250190449 A1; Zhang; Bo et al. (hereinafter Zhang)
Regarding claim 1, Arriaga teaches A method comprising: identifying, by a processing system including at least one processor, a user of a text-to-query written in structured query language programming language system; (Arriaga [FIG.1] shows the system identifying a user of a text-to-query written in structured query language programming language system[0004] The memory instructs the processor to generate one or more target profiles as function of the modified dataset, the one or more protection gaps, and a user input. The memory instructs the processor to generate a video report as a function of the one or more target profiles. [0068] natural language processing tools and AI-driven response systems that allow the avatar to interpret user inputs accurately and respond appropriately in real-time. [138-143] elaborate on the matter) identifying, by the processing system, a machine learning model that has been trained based on at least one preference of the user to predict a presence of a target feature …(Arriaga [0022] With continued reference to FIG. 1, computing device 104 may perform determinations, classification, and/or analysis steps, methods, processes, or the like as described in this disclosure using machine-learning processes. A “machine-learning process,” as used in this disclosure, is a process that automatedly uses a body of data known as “training data” and/or a “training set” (described further below in this disclosure) to generate an algorithm that will be performed by a Processor module to produce outputs given data provided as inputs; this is in contrast to a non-machine learning software program where the commands to be executed are determined in advance by a user and written in a programming language. A machine-learning process may utilize supervised, unsupervised, lazy-learning processes and/or neural networks, described further below. [0061] one or more gap finder questions based on the presence of one or more elements categorized to a particular protection categorization 172. For example, gap finder module 156 may be configured to ask questions associated with vehicle coverage when at least one element within target data 124 and/or modified dataset 144 is associated to a vehicle protection categorization 172. In one or more embodiments, the chatbot may be configured to define terms for a user and/or target. For example, a target may input a query into chatbot asking what a particular term means wherein chatbot may be configured to define the terms. In one or more embodiments, the chatbot may provide visual aid, tool tips, and videos in order to define complex terms and help an individual understand one or more terms. In one or more embodiments, chatbot may utilize a machine learning model, such as any machine learning model as described in this disclosure wherein training data may be used to generate outputs that can educate individuals on complex terms. In one or more embodiments, training data may include previous inputs by previous individuals into the chatbot and outputs that provide visual aid, videos, and the like. In an embodiment, each input by a user into chatbot may be used to train the machine learning model wherein responses such as “I still don't understand” by the individual may indicate that a particular term requires more a more precise definition. In one or more embodiments, user input into chatbot system may be used to train the machine learning model wherein chatbot may be configured to provide more accurate results on every iteration. In one or more embodiments, training of the machine learning model, and as a result training chatbot, may allow for quicker and more efficient communication between chatbot and an individual. In an embodiment, training data may allow for shorter communications with chatbot, and as result, less strain on one or more computing devices configured to generate chatbot.[0129] With continued reference to FIG. 1, apparatus 100 may include a unified dashboard, wherein the unified dashboard may include a predictive model, wherein memory 112 may contain instructions configuring at least a processor 108 to send the user a notification based on an output of the predictive model. As used in this disclosure, a “unified dashboard” is a control panel within a graphical user interface. In a non-limiting example, the unified dashboard may utilize a recommendation model. As used in this disclosure, a “recommendation model” is an algorithm or machine learning model designed to make relevant suggestions to users based on their preferences... model may be trained on user interaction data, such as, without limitation, past conversation history correlated to user preferences and behaviors...[137-145] elaborates on the mater [FIG.1 in conjunction with FIG.4] shows corresponding visual) receiving, by the processing system from the user, a question phrased in natural language; (Arriaga [FIG.1] shows receiving the user input/question [0039] With continued reference to FIG. 1, dataset 120 and/or elements thereof may be received by a chatbot system. A “chatbot system” for the purposes of this disclosure is a program configured to simulate human interaction with a user with a user in order to receive or convey information. In some cases, chatbot system may be configured to receive dataset 120 and/or elements thereof through interactive questions presented to the user. the questions may include, but are not limited to, questions such as “What is your name?,” “What is your date of birth?”, “Please list any assets owned having a value of above 1,000$?” and the like. In some cases, computing device 104 may be configured to present a comment box through a user interface wherein a user may interact with the chatbot and answer the questions through input into the chat box. In some cases, questions may require selection of one or more pre-configured answers. For example, chatbot system may ask a user to select the appropriate salary range corresponding to the user, wherein the user may select the appropriate range from a list of pre-configured answers. In situations where answers are limited to limited responses, chatbot may be configured to display checkboxes wherein a user may select a box that is most associated with their answer. In some cases, chatbot may be configured to receive dataset 120 or target data 124 and through an input. In some cases, each question may be assigned to a particular categorization wherein a response to the question may be assigned to the same categorization. For example, a question prompting a user to input an income may be assigned to an income categorization wherein a response from the user may also be assigned to the income categorization. [0060] receive a response from a user of one or more elements that have been classified and may require protection based on one or more lookup tables. For example, processor 108 may generate a question to ask a user if a particular vehicle that has been classified currently contains insurance coverage. If the user answers yes, then processor 108 does not generate a protection gap 152. If however, the user inputs 168 that the particular element does not have protection, then processor 108 may generate a protection gap 152. [61-69] elaborate on the matter) generating, by the processing system using a language model, a query written in structured query language programming language designed to retrieve an answer to the question; (Arriaga [0037] The instant solution may be implemented in conjunction with computing environments involving technology classified under one or more of the artificial intelligence (AI) classifications and/or AI models[0047] In FIG. 2, the AI engine 222 may control access to models that are stored within the model repository 223. For example, the models may include AI models, generative AI models, machine learning models, neural networks, LLMs, and/or the like. The software application 210 may trigger execution of the model 224 from the model repository 223 by invoking an API 221 of the AI engine 222. The request may include an identifier (ID) of the model 224 such as a unique ID assigned by the host platform 220, a payload of data (e.g., to be input to the model during execution), and the like. The AI engine 222 may retrieve the model 224 from the model repository 223 in response and deploy the model 224 within a live runtime environment. After the model is deployed, the AI engine 222 may execute the running instance of the model 224 on the payload of data and return a result of the execution to the software application 210 [0055] machine-learning module may obtain a training set by querying a communicatively connected database that includes past inputs and outputs. Training data may include inputs from various types of databases, resources, and/or user inputs and outputs correlated to each of those inputs so that a machine-learning model may determine an output. Correlations may indicate causative and/or predictive links between data, which may be modeled as relationships, such as mathematical relationships, by machine-learning models, as described in further detail below. In one or more embodiments, training data may be formatted and/or organized by categories of data elements by, for example, associating data elements with one or more descriptors corresponding to categories of data elements. As a non-limiting example, training data may include data entered in standardized forms by persons or processes, such that entry of a given data element in a given field in a form may be mapped to one or more descriptors of categories. Elements in training data may be linked to categories by tags, tokens, or other data elements [0056] machine-learning module may obtain a training set by querying a communicatively connected database that includes past inputs and outputs [55-60] further elaborate [FIG.1] shows using a language model to make a query written in structured query language programming language designed to retrieve an answer to the question) extracting, by the processing system, a set of query features and a set of target features from the query written in structured query language programming language; (Arriaga [0019] the present disclosure can be used to parse through datasets and determine a validity status of the elements within the datasets. Aspects of the present disclosure can also be used to generate target profiles for a particular target. Exemplary embodiments illustrating aspects of the present disclosure are described below in the context of several specific examples.[0035] Still referring to FIG. 1, in some embodiments, an OCR process may include a feature extraction process...machine-learning process like nearest neighbor classifiers (e.g., k-nearest neighbors algorithm) can be used to compare image features with stored glyph features and choose a nearest match. OCR may employ any machine-learning process described in this disclosure, for example machine-learning processes described with reference to FIGS. 4-6. Exemplary non-limiting OCR software includes Cuneiform and Tesseract. Cuneiform is a multi-language, open-source optical character recognition system originally developed by Cognitive Technologies[0047] With continued reference to FIG. 1, computing device 104 and/or processor 108 may be configured to generate classifiers as described throughout this disclosure using a K-nearest neighbors (KNN) algorithm. A “K-nearest neighbors algorithm” as used in this disclosure, includes a classification method that utilizes feature similarity to analyze how closely out-of-sample-features resemble training data to classify input data to one or more clusters and/or categories of features as represented in training data; this may be performed by representing both training data and input data in vector forms, and using one or more measures of vector similarity to identify classifications within training data, and to determine a classification of input data. ... [88-98] goes into further detail on the matter [FIG.1] shows corresponding visual) generating, by the processing system using the machine learning model and based on the set of query features and the set of target features, a prompt that makes a recommendation with respect to the presence of the target feature in the query written in structured query language programming language; (Arriaga [0039]chatbot may be configured to receive dataset 120 or target data 124 and through an input. In some cases, each question may be assigned to a particular categorization wherein a response to the question may be assigned to the same categorization. For example, a question prompting a user to input an income may be assigned to an income categorization wherein a response from the user may also be assigned to the income categorization.[0061] With continued reference to FIG. 1, gap finder module 156 may further include a chatbot configured to ask one or more gap finder questions. “Gap finder questions” for the purposes of this disclosure are a series of questions or statements wherein a particular response to the questions may indicate one or more protection gaps 152 associated with the target. [0063] target has, and/or personalized advice received from one or more ally's and/or insurance agents. In one or more embodiments, processor may utilize chatbot as described above to receive prompts regarding any elements within stewardship file 180 wherein chatbot may be configured to define terms and assist within one or more generated elements within stewardship file 180. In some cases, stewardship file 180 may include a personalized, digital (including video) report containing an update of how current trends are affecting their insurance situation, highlights of their insurance program, their 5-year claim and motor vehicle history, premium analysis, and recommendations for their insurance renewal. In one or more embodiments, stewardship file 180 may provide an engaging insurance experience while preventing coverage gaps from occurring. In one or more embodiments, stewardship file 180 may include modification and/or updates periodically based on changes in the target's situation as indicated by an input of data, by a modification of target data and/or market conditions. In one or more embodiments, processor 108 may be configured to generate personalized alerts based on elements within stewardship file 180 and/or updates made to stewardship file 180. For example, if stewardship file 180 identifies a potential coverage gap, the target or an individual associated with stewardship file 180 may receive an alert with a recommendation [103-108] elaborate on the matter [129] the machine learning model used for the recommendation model may be trained on user interaction data, such as, without limitation, past conversation history correlated to user preferences and behaviors. Continuing, the previous non-limiting example, the machine learning model may include any machine learning model described herein. Refer to FIGS. 4-6 for a more detailed description of the machine learning model. In another non-limiting example, the recommendation model may be trained on user interaction data such as clickstream data, time spent on different sections of the dashboard, and user feedback. Continuing, without limitation, the recommendation model may be trained on historical data such as past interactions with similar systems, previous purchases, and user preferences. Continuing, without limitation, the recommendation model [FIG.4] shows a visual) providing, by the processing system, the prompt to the language model to cause the language model to refine the query written in structured query language programming language in response to the recommendation; (Arriaga [0063] target has, and/or personalized advice received from one or more ally's and/or insurance agents. In one or more embodiments, processor may utilize chatbot as described above to receive prompts regarding any elements within stewardship file 180 wherein chatbot may be configured to define terms and assist within one or more generated elements within stewardship file 180. In some cases, stewardship file 180 may include a personalized, digital (including video) report containing an update of how current trends are affecting their insurance situation, highlights of their insurance program, their 5-year claim and motor vehicle history, premium analysis, and recommendations for their insurance renewal. In one or more embodiments, stewardship file 180 may provide an engaging insurance experience while preventing coverage gaps from occurring. In one or more embodiments, stewardship file 180 may include modification and/or updates periodically based on changes in the target's situation as indicated by an input of data, by a modification of target data and/or market conditions. In one or more embodiments, processor 108 may be configured to generate personalized alerts based on elements within stewardship file 180 and/or updates made to stewardship file 180. For example, if stewardship file 180 identifies a potential coverage gap, the target or an individual associated with stewardship file 180 may receive an alert with a recommendation [103-108] elaborate on the matter [129] ... the recommendation model may be trained on user demographic data, such as age, location, income level, and other relevant demographic information. Continuing, without limitation, the recommendation model may be trained on user behavioral data, such as, patterns of behavior, for example, frequently viewed content, preferred communication channels, and response to previous recommendations, and the like. Continuing, without limitation, the recommendation model may be trained on feedback data, such as, user ratings and reviews of past recommendations, which can be used to refine the model's accuracy. In a non-limiting example, the recommendation model may use the training data to recommend the user insurance products, risk management strategies, value-added services, educational resources, promotions and discounts, and the like. Insurance products: Suggesting new or additional insurance policies that align with the user's needs and preferences. Continuing, without limitation, the recommendation model may recommend...[FIG.4] shows a visual) and querying, by the processing system, a database using the query written in structured query language programming language that has been refined. (Arriaga [FIG.1 in conjunction with FIG.4] show querying a database using the query written in structured query language programming language that has been refined [133] In some embodiments, the web crawler may be trained with information received from a user through a graphical user interface. In some embodiments, the web crawler may be configured to generate a web query. For example, without limitation, a web query may include search criteria received from a user. For example, a user may submit a plurality of websites for the web crawler to search to extract entity records, inventory records, pricing records, product records, customer records, financial transaction records, customer feedback and review records, and the like. In a non-limiting example, the personalized content feed may recommend articles based on target profile 176 by classifying target profile 176 into cohorts using a classification model. For example, without limitation, target profile 176 may be classified into cohorts such as “young professionals,” “families,” retirees,” and the like. In another non-limiting example the classification model may analyze user preferences, behaviors, interest, interactions with the platform, and the like to determine the cohorts and assign target profile 176 to a specific cohort. In a non-limiting example, the classification model may include one or more algorithms and/or machine learning models to analyze and classify target profiles 176. In a non-limiting example, the personalized content feed may be integrated into the unified dashboard. Without limitation, the personalized content feed may interactively and continuously learn and adjust recommendations based on the latest user behaviors, interactions, interests, and the like to ensure that the personalized content feed remains relevant and useful to the user. In some embodiments, training data for the classification model may include exemplary target profiles correlated to cohort labels. [0137] With continued reference to FIG. 1, apparatus 100 may include a summary generator, wherein the summary generator may include a large language model configured to receive the plurality of target data 124 as input and output a summary of the plurality of target data 124. In some embodiments, large language model may be configured to receive target data 124 as part of a natural language prompt. For example, the prompt could include: “Please generate a summary of this target data: [Target data].” In some embodiments, large language model may be configured to automatically retrieve target data 124 and generate a summary of the target data 124 as a function of a natural language prompt. For example, prompt may include “please generate a summary for [target] and LLM may automatically retrieve target data 124 as a function of the identification of the target. [133-140] elaborate on the matter) Arriaga lacks explicitly and orderly teaching methods and corresponding processes based on data in an input query written in structured query language programming language based on a presence of a query feature in the input query written in structured query language programming language; wherein the machine learning model has been trained using a signature table that captures user preferences with respect to a plurality of test pairs However Zhang teaches corresponding processes based on data in an input query written in structured query language programming language based on a presence of a query feature in the input query written in structured query language programming language; wherein the machine learning model has been trained using a signature table that captures user preferences with respect to a plurality of test pairs (Zhang [0005] analyze the structured data or to be trained in the use of a particular application interface. The system may also be configured to perform continuous learning such that user feedback and preferences may be provided based on the response, and the system may adjust parameters of the prompt processing and agent selection processes to increase accuracy and user-satisfaction with the generative AI-assisted analytics performed by the system. [0031] The structured data set 154 may include multiple file types or a single file type, and may include or be organized according to multiple schemas, multiple data table formats, multiple data types, or a combination thereof. [0040]The generative AI agents 126 include any types of generative AI agents that are capable of performing analytics or mathematical operations on input data, such as a comma-separated value (CSV) agent, a structured query language (SQL) ... [0040] AI models (referred to herein collectively as “the AI models 125”) that are trained to perform the tasks described with reference to the agent orchestrator 124. For example, the AI models 125 may include a trained AI classifier or ML classifier that is configured to select one or more of the generative AI agents 126 based on the prompt 170 and the structured data 172. [0044] user feedback 180 that indicates the user's feedback and any preferences associated with the prompt 170 and the response 110, the GUI 174, and/or the visual elements 176. For example, the user feedback 180 may indicate an accuracy score and a preference for a line graph that is related to a user prompt that requested performance of some statistical analysis and “a graph displaying the results.” The computing device 102 may modify the knowledge graph 120 based on the prompt 170, the response 110, and the user feedback 180. For example, one or more nodes may be added to the knowledge graph 120 based on preferences indicated by the user feedback [0056] sub-intents related to user prompts and responses in order to facilitate adaptation to user feedback of responses output by the generative AI analytics engine 200. In such an example, a received user feedback that indicates a user's feedback and preferences related to a response may be used to update weights and relationships in the knowledge graph in real-time [90-96] elaborate on the matter of using a signature table that captures user preferences with respect to a plurality of test pairs [FIG.1 in conjunction with FIG.3] show actions and processes based on an input query written in structured query language programming language based on a presence of a query feature in the input query written in structured query language programming language; wherein the machine learning model has been trained using a signature table that captures user preferences with respect to a plurality of test pairs) Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to take all prior methods and make the addition of Zhang in order to help efficiently improve the querying output of the system (Zhang [0003] Additionally, these analytical applications are typically designed to with the functionality as the primary concern, often without prioritizing ease of use, particularly by customers who are inexperienced in using data-driven user interfaces. As technology has advanced, artificial intelligence and machine learning has been leveraged to create generative artificial intelligence models, such as large language models (LLMs), that can create novel text outputs that can be used to improve ease of use and user interactivity with user interfaces of applications. However, LLMs are typically not suited for performing analytics and can be prone to “hallucinations,” such as generating incorrect information in response to a question for which the LLM is unable to correctly answer. As such, although LLMs and generative artificial intelligence have improved usability of some applications, these improvements have not extended to analytical applications and to structured data set contexts.[0004] prompt optimization operations, or a combination thereof, to modify the prompt such that the prompt is more useful to the agent orchestrator (e.g., so that performance of the agent orchestrator is improved as compared to using the unmodified user prompt). The agent orchestrator may include a trained AI classifier configured to select the one or more generative AI agents from multiple available generative AI agents that are configured to perform analytics tasks on numerical and structured data. As non-limiting examples, the generative AI agents may include a CSV agent, an SQL agent, a Spark agent, a semantic search agent, or the like. The agent orchestrator may select the generative AI agents based on agent features associated with the available generative AI agents, intent features associated with the intent determined from the user prompt, data schema features associated with the structured data set, or a combination thereof, and optionally based on a greedy parameter to enable exploration of agents with lower scores based on current knowledge but that may have the potential to yield better results.[0005] a table (or portion thereof), or other visual elements in addition to text output and/or numerical output to provide a user with information to enable efficient derivation of insights and decision-making capabilities. Such an output may have greater utility than a wholly text or numerical output, thereby improving a user experience and enabling easier and more accurate decision-making and understanding of insights unlocked from structured data without requiring the user to manually analyze the structured data or to be trained in the use of a particular application interface. The system may also be configured to perform continuous learning such that user feedback and preferences may be provided based on the response, and the system may adjust parameters of the prompt processing and agent selection processes to increase accuracy and user-satisfaction with the generative AI-assisted analytics performed by the system.[0031] the prompt 170 and/or the structured data set 154 and to use the agent orchestrator 124 to select one or more of the generative AI agents 126 for providing an accurate and efficient response to the prompt 170, as further described herein [0043] answer the question(s) included in the prompt 170 to the user of the user device 150, to enable quick and efficient understanding by the user, thereby enabling fast and effective decision making based on insights drawn from the structured data set [FIG.1] shows overall visual)
Corresponding product claim 12 is rejected similarly as claim 1 above. Additional Limitations: computer readable medium capable of reading and executing instructions (Arriaga [FIG.1] shows corresponding with computer readable medium capable of reading and executing instructions[0182] Memory 908 may include various components (e.g., machine-readable media) including, but not limited to, a random-access memory component, a read only component, and any combinations thereof. In one example, a basic input/output system 916 (BIOS), including basic routines that help to transfer information between elements within computer system 900, such as during start-up, may be stored in memory 908. Memory 908 may also include (e.g., stored on one or more machine-readable media) instructions (e.g., software) 920 embodying any one or more of the aspects and/or methodologies of the present disclosure. In another example, memory 908 may further include any number of program modules including, but not limited to, an operating system, one or more application programs, other program modules, program data, and any combinations thereof[183] storage device 924 and an associated machine-readable medium 928 may provide nonvolatile and/or volatile storage of machine-readable instructions, data structures, program modules, and/or other data for computer system 900. In one example, software 920 may reside, completely or partially, within machine-readable medium 928. In another example, software 920 may reside, completely or partially, within processor 904. ) [177-182] further elaborate on the matter)
Regarding claim 2, Arriaga and Zhang teach The method of claim 1, wherein the question includes at least one term that is capable of being interpreted in more than one different way. (Arriaga [0053] Additionally, or alternatively, rule-based engine may include an inference engine to determine a match of protection rule, where any or all elements within modified dataset 144 may be represented as values for linguistic variables measuring the same. In some cases, each rule within protection rule may include a rule and a corresponding action associated with the rule. In some cases, protection rule may include a rule such as “if the asset does is not fully covered under an insurance policy” and a corresponding action indicating “generate a protection indicating that an asset is not fully covered under the insurance policy.” In some cases, inference engine may be configured to determine which rule out of a plurality of rules should be executed with respect to a particular element within modified dataset 144. For example, inference engine may determine that a particular rule relating to policy limits should be selected when the elements within modified dataset 144 discuss a particular policy limit. Similarly, a particular rule relating to types of protection may be selected when elements within modified dataset 144 indicate types of protection. In some cases, gap finder module 156 may receive elements within modified dataset 144 and/or target data [0070] With continued reference to FIG. 1, Text-to-speech (TTS)...Once the script is ready, the TTS system may analyze the text to phonetically understand and process the words. In some cases, TTS systems may employ natural language processing algorithms to interpret the context of the dialogue, ensuring that the intonation and emphasis are appropriate for the content's sentiment and importance. This means that the avatar can dynamically adjust its tone, speed, and expressiveness based on the script's cues, making the conversation feel more natural. Additionally, TTS technology can support multiple languages and accents, offering a wide range of voices from which to choose, thus aligning with the demographic and personal preferences of the user. For instance, if a script includes specific guidance on proceeding with a claim after a natural disaster, the TTS system can deliver this information with the seriousness and urgency it requires, while also providing comforting reassurance through tone adjustments. Integrating TTS into the digital avatar's operational framework allows for real-time dialogue delivery where the avatar not only speaks the information but can also respond to user inputs or changes in conversation flow, as dictated by the script's decision trees or conditional logic. [101] . Still referring to FIG. 1, the target machine learning model may include a large language model (LLM). A “large language model,” as used herein, is a deep learning data structure that can recognize, summarize, translate, predict and/or generate text and other content based on knowledge gained from massive datasets. [134-139] elaborates on the prediction methods (search that is capable of being interpreted in more than one different way.))
Regarding claim 3, Arriaga and Zhang teach The method of claim 1, wherein the machine learning model is one of a plurality of user-specific machine learning models stored in a second database that is accessible to the processing system. (Arriaga [0055] With continued reference to FIG. 1, gap finder module 156 may include a protection machine learning model 160. Processor 108 and/or gap finder module 156 may use a machine learning module, such as a protection machine learning module for the purposes of this disclosure, to implement one or more algorithms or generate one or more machine-learning models, such as a protection machine learning model 160, to calculate at least one protection gap 152. However, the machine learning module is exemplary and may not be necessary to generate one or more machine learning models and perform any machine learning described herein. In one or more embodiments, one or more machine-learning models [0087] With continued reference to FIG. 1, in one or more embodiments, processor 108 may implement one or more aspects of “generative artificial intelligence (AI),” a type of AI that uses machine learning algorithms to create, establish, or otherwise generate data such as, without limitation, stewardship file, video report 178. and/or the like in any data structure as described herein (e.g., text, image, video, audio, among others) that is similar to one or more provided training examples. In an embodiment, machine learning module described herein may generate one or more generative machine learning models that are trained on one or more set of target training data. One or more generative machine learning models may be configured to generate new examples that are similar to the training data of the one or more generative machine learning models but are not exact replicas; for instance, and without limitation, data quality or attributes of the generated examples may bear a resemblance to the training data provided to one or more generative machine learning models, wherein the resemblance may pertain to underlying patterns, features, or structures found within the provided training data. [0088] Still referring to FIG. 1, in some cases, generative machine learning models may include one or more generative models. As described herein, “generative models” refers to statistical models of the joint probability distribution P(X,Y) on a given observable variable x, representing features or data that can be directly measured or observed (e.g., target profiles 176) and target variable y, representing the outcomes or labels that one or more generative models [89-98] goes into detail on wherein the machine learning model is one of a plurality of user-specific machine learning models stored in a second database that is accessible to the processing system)
Regarding claim 4, Arriaga and Zhang teach The method of claim 1, wherein the set of query features includes at least one of: a table referenced by the query written in structured query language programming language, a number of subqueries contained in the query written in structured query language programming language, a number of joins contained in the query written in structured query language programming language, a number of where clauses contained in the query written in structured query language programming language, a number of fields of the table returned by the query written in structured query language programming language, whether a top-level query written in structured query language programming language performs aggregation, a type of field in the query written in structured query language programming language, an order by clause in the query written in structured query language programming language, a null check in the query written in structured query language programming language, an aggregation in the query written in structured query language programming language, or a having clause in the query written in structured query language programming language. (Arriaga [0108] With continued reference to FIG. 1, multi-headed attention in encoder may apply a specific attention mechanism called self-attention. Self-attention allows models such as an LLM or components thereof to associate each word in the input, to other words. As a non-limiting example, an LLM may learn to associate the word “you”, with “how” and “are”. It's also possible that an LLM learns that words structured in this pattern are typically a question and to respond appropriately. In some embodiments, to achieve self-attention, input may be fed into three distinct fully connected neural network layers to create query, key, and value vectors. A query vector may include an entity's learned representation for comparison to determine attention score. A key vector may include an entity's learned representation for determining the entity's relevance and attention weight. A value vector may include data used to generate output representations. Query, key, and value vectors may be fed through a linear layer; then, the query and key vectors may be multiplied using dot product matrix multiplication in order to produce a score matrix. The score matrix may determine the amount of focus for a word should be put on other words (thus, each word may be a score that corresponds to other words in the time-step). [0109] Still referencing FIG. 1, in order to use self-attention in a multi-headed attention computation, query, key, and value may be split into N vectors before applying self-attention. Each self-attention process may be called a “head.” Each head may produce an output vector and each output vector from each head may be concatenated into a single vector. This single vector may then be fed through the final linear layer discussed above. In theory, each head can learn something different from the input, therefore giving the encoder model more representation power. [138] the API query may include authentication information such as API keys, tokens, or other credentials required to authenticate the request with the third-party application. In another non-limiting example, the API query may include the type of request being made, such as “GET,” “POST,” “PUT,” and the like, depending on the action required. For example, without limitation the API may query to an insurance carrier's APU to retrieve updated policy information. [56-61] further elaborate on the query features)
Regarding claim 5, Arriaga and Zhang teach The method of claim 1, wherein the set of target features includes at least one of: whether a top-level query written in structured query language programming language uses select distinct, whether the query written in structured query language programming language uses an is not null predicate, whether the query written in structured query language programming language uses a null check, whether the query written in structured query language programming language uses an order by that involves a constraint on a subquery, whether the query written in structured query language programming language uses a join to table which does not contribute any fields to an output, a number of fields in the query written in structured query language programming language, or a data manipulation in the query written in structured query language programming language. (Arriaga [0040] With continued reference to FIG. 1, dataset 120 may be retrieved using a web crawler. A “web crawler,” as used herein, is a program that systematically browses the internet for the purpose of Web indexing. The web crawler may be seeded with platform URLs, wherein the crawler may then visit the next related URL, retrieve the content, index the content, and/or measures the relevance of the content to the topic of interest. In some embodiments, computing device 104 may generate a web crawler to compile dataset 120. The web crawler may be seeded and/or trained with a reputable website, such as government websites. A web crawler may be generated by computing device 104. In some embodiments, the web crawler may be trained with information received from a user through a user interface. In some embodiments, the web crawler may be configured to generate a web query. A web query may include search criteria received from a user. For example, a user may submit a plurality of websites for the web crawler to search to extract any data suitable for dataset 120[0119] With continued reference to FIG. 1, an LLM may generate at least one annotation as an output. At least one annotation may be any annotation as described herein. In some embodiments, an LLM may include multiple sets of transformer architecture as described above. Output may include a textual output. A “textual output,” for the purposes of this disclosure is an output comprising a string of one or more characters. Textual output may include, for example, a plurality of annotations for unstructured data. In some embodiments, textual output may include a phrase or sentence identifying the status of a user query. In some embodiments, textual output may include a sentence or plurality of sentences describing a response to a user query. As a non-limiting example, this may include restrictions, timing, advice, dangers, benefits, and the like...[133] the personalized content feed may interactively and continuously learn and adjust recommendations based on the latest user behaviors, interactions, interests, and the like to ensure that the personalized content feed remains relevant and useful to the user. In some embodiments, training data for the classification model may include exemplary target profiles correlated to cohort labels. [138] API query may include authentication information such as API keys, tokens, or other credentials required to authenticate the request with the third-party application. In another non-limiting example, the API query may include the type of request being made, such as “GET,” “POST,” “PUT,” and the like, depending on the action required. For example, without limitation the API may query to an insurance carrier's APU to retrieve updated policy information. Continuing the previous non-limiting example, apparatus 100 may receive a response from the API containing the requested data, such as, policy details, quote adjustments, claim information, and the like. In another non-limiting example, the API layer may be configured to facilitate the integration of third-party applications. As used in this disclosure, a “third-party application” is a software application developed by an entity other than the primary system vendor or integrator. In some cases, third-party applications may include additional, non-essential functions and may not be part of core system software. In some cases, third-party application may require a specific runtime environment to function known as the “proprietary runtime environment.” In some cases, proprietary runtime environment may include one of more libraries, services, or other dependencies that are unique to applications, and not necessarily shared with other parts of the system. For example, without limitation, the third-party application may include an insurance carrier website, wherein application 100 allows single-click access to change the user's policy or access the client's documents. In another non-limiting example, the API layer may integrate with the third-party application, such as a third-party home replacement cost estimation tool within the platform. Continuing the previous non-limiting example, integration with the third-party home replacement cost estimation tool may allow agency staff to easily access, modify, and update [133-138] elaborates on the query features)
Regarding claim 6, Arriaga and Zhang teach The method of claim 1, wherein the machine learning model takes the set of query features and the set of target features as inputs and generates as an output a prediction as to whether the target feature should be added to the query written in structured query language programming language or removed from the query written in structured query language programming language. (Arriaga [0053] include dataset 120, target data .... inference engine may determine that a particular rule relating to policy limits should be selected when the elements within modified dataset 144 discuss a particular policy limit. Similarly, a particular rule relating to types of protection may be selected when elements within modified dataset 144 indicate types of protection. In some cases, gap finder module 156 may receive elements within modified dataset 144 and/or target data 124 and make calculations ... within target data 124. For example, web crawler may be configured to retrieve an estimate of the target's property using estimates from one or more property websites. [0101] Still referring to FIG. 1, the target machine learning model may include a large language model (LLM). A “large language model,” as used herein, is a deep learning data structure that can recognize, summarize, translate, predict and/or generate text and other content based on knowledge gained from massive datasets. Large language models may be trained on large sets of data. Training sets may be drawn from diverse sets of data such as, as non-limiting examples, novels, blog posts, articles, emails, unstructured data, electronic records, and the like. In some embodiments, training sets may include a variety of subject matters, such as, as nonlimiting examples, weather reports, insurance policies, insurance claims, property damage reports, emails, user communications, advertising documents, newspaper articles, and the like. In some embodiments, training sets of an LLM may include information from one or more public or private databases. As a non-limiting example, training sets may include databases associated with an entity. In some embodiments, training sets may include portions of documents associated with the electronic records 112 correlated to examples of outputs. In an embodiment, an LLM may include one or more architectures based on capability requirements of an LLM. Exemplary architectures may include, without limitation, GPT (Generative Pretrained Transformer), BERT (Bidirectional Encoder Representations from Transformers), T5 (Text-To-Text Transfer Transformer), and the like. Architecture choice may depend on a needed capability such generative, contextual, or other specific capabilities.) [102] A model such as an LLM may learn by adjusting its parameters during the training process to minimize a defined loss function, which measures the difference between predicted outputs and ground truth. Once a model has been generally trained, the model may then be specifically trained to fine-tune the pretrained model on task-specific data to adapt it to the target task. Fine-tuning may involve training a model with task-specific training data, adjusting the model's weights [107,116 and 151] elaborate on the matter)
Regarding claim 7, Arriaga and Zhang teach The method of claim 6, wherein the language model adds the target feature to the query written in structured query language programming language prior to the querying, in response to the prompt. (Arriaga [0063] within target data 124 and/or modified dataset 144 associated with one or more protection gaps 152. In some cases, stewardship file 180 may include one or more protection gaps 152 determined above. In some cases, each protection gap 152 may contain a corresponding insurance plan wherein stewardship file 180 may include the corresponding insurance plans. In some cases, stewardship file 180 may include one or more images of elements associated with assets described within the target data 120 and the insurance plan associated with the images. For example, stewardship file 180 may include an image of a vehicle and a corresponding insurance coverage. In one or more embodiments, stewardship file 180 may include updates on particular trends within a geographic area, any current insurance claims the target has, and/or personalized advice received from one or more ally's and/or insurance agents. In one or more embodiments, processor may utilize chatbot as described above to receive prompts regarding any elements within stewardship file 180 wherein chatbot may be configured to define terms and assist within one or more generated elements within stewardship file 180. In some cases, stewardship file 180 may include a personalized, digital (including video) report containing an update of how current trends are affecting their insurance situation, highlights of their insurance program, their 5-year claim and motor vehicle history, premium analysis, and recommendations for their insurance renewal. In one or more embodiments, stewardship file 180 may provide an engaging insurance experience while preventing coverage gaps from occurring. In one or more embodiments, stewardship file 180 may include modification and/or updates periodically based on changes in the target's situation as indicated by an input of data, by a modification of target data and/or market conditions. In one or more embodiments, processor 108 may be configured to generate personalized alerts based on elements within stewardship file 180 and/or updates made to stewardship file 180. For example, if stewardship file 180 identifies a potential coverage gap, the target or an individual associated with stewardship file 180 may receive an alert with a recommendation on how to address it. In one or more embodiments, processor may be configured to receive updated situations associated with a target, such as updated trends, changes to target data 124 and the like.[0129] With continued reference to FIG. 1, apparatus 100 may include a unified dashboard, wherein the unified dashboard may include a predictive model, wherein memory 112 may contain instructions configuring at least a processor 108 to send the user a notification based on an output of the predictive model. As used in this disclosure, a “unified dashboard” is a control panel within a graphical user interface. In a non-limiting example, the unified dashboard may utilize a recommendation model. As used in this disclosure, a “recommendation model” is an algorithm or machine learning model designed to make relevant suggestions to users based on their preferences, behaviors, and interactions. In a non-limiting example, the recommendation model may personalize content within the digital environment wherein the personalized content may enable the user to discover items of interest. In a non-limiting example, the recommendation model may suggest personalized risk management strategies, resources, insurance products, and value-added services based on the customer's target profile [53-63] elaborate on the matter)
Regarding claim 8, Arriaga and Zhang teach The method of claim 6, wherein the language model removes the target feature from the query written in structured query language programming language prior to the querying, in response to the prompt. (Arriage [0046] With continued reference to FIG. 1, in some embodiments, classifier training data may be iteratively updated using feedback. Feedback, in some embodiments, may include user feedback. For example, user feedback may include a rating, such as a rating from 1-10, 1-100, −1 to 1, “happy,” “sad,” and the like. In some embodiments, user feedback may rate a user's satisfaction with the target categorization. In some embodiments, feedback may include outcome data. “Outcome data,” for the purposes of this disclosure, is data including an outcome of a process. As a non-limiting example, outcome data may include information regarding whether a target made a purchase, whether a target has continued communications, and the like. Iteratively updating classifier training data may include removing datasets and target categorizations from classifier training data as a function of negative or unfavorable feedback. [0051] With continued reference to FIG. 1, processor 108 is configured to modify the dataset 120 as a function of the validity status 136. Processor 108 may modify dataset 120 by removing any target data 124 that may be considered invalid based on validity status 136. For example, processor 108 may be configured to remove target data [0053] target data 124 and/or any other data described in this disclosure. In some embodiments, protection rule may be stored in a database as described in this disclosure. Additionally, or alternatively, rule-based engine may include an inference engine to determine a match of protection rule, where any or all elements within modified dataset 144 may be represented as values for linguistic variables measuring the same. In some cases, each rule within protection rule may include a rule and a corresponding action associated with the rule. In some cases, protection rule may include a rule such as “if the asset does is not fully covered under an insurance policy” and a corresponding action indicating “generate a protection indicating that an asset is not fully covered under the insurance policy.” In some cases, inference engine may be configured to determine which rule out of a plurality of rules should be executed with respect to a particular element within modified dataset 144. For example, inference engine may determine that a particular rule relating to policy limits should be selected when the elements within modified dataset 144 discuss a particular policy limit. Similarly, a particular rule relating to types of protection may be selected when elements within modified dataset 144 indicate types of protection. In some cases, gap finder module 156 may receive elements within modified dataset 144 and/or target data 124 and make calculations [102] A model such as an LLM may learn by adjusting its parameters during the training process to minimize a defined loss function, which measures the difference between predicted outputs and ground truth. Once a model has been generally trained, the model may then be specifically trained to fine-tune the pretrained model on task-specific data to adapt it to the target task. Fine-tuning may involve training a model with task-specific training data, adjusting the model's weights to optimize performance for the particular task. In some cases, this may include optimizing the model's performance by fine-tuning hyperparameters such as learning rate, batch size, and regularization. Hyperparameter tuning may help in achieving the best performance and convergence during training. In an embodiment, fine-tuning a pretrained model such as an LLM may include fine-tuning the pretrained model using Low-Rank Adaptation (LoRA). As used in this disclosure, “Low-Rank Adaptation” is a training technique for large language models that modifies a subset of parameters in the model. Low-Rank Adaptation may be configured to make the training process more computationally efficient by avoiding a need to train an entire model from scratch. In an exemplary embodiment, a subset of parameters that are updated may include parameters that are associated with a specific task or domain.)
Regarding claim 10, Arriaga and Zhang teach The method of claim 1, wherein each test pair of the plurality of test pairs comprises: a test question phrased in natural language and a machine learning model-generated query written in structured query language programming language corresponding to the test question. (Arriaga [0072] The processor may utilize natural language processing (NLP) techniques to understand the context and significance of the text, allowing it to structure a script that logically sequences the information in a way that's easy for users to follow. Once the script is prepared, Processor 108 may segment the script into discrete sections corresponding to different aspects of the insurance information, such as coverage details, risk assessments, and procedural guides. Each segment may then be assigned specific visual and auditory features that will complement and enhance the textual content. For example, segments explaining complex insurance terms might be paired with visual aids like charts or diagrams, [134] In a non-limiting example, the trained predictive model may be validated and tested using a separate dataset to evaluate its accuracy and performance. Continuing the previous non-limiting example, validation and texting of the predictive model may ensure that the model can generalize well to new, unseen data. In a non-limiting example, the predictive model may continuously monitor the incoming data. Continuing, the predictive model may detect potential risks based on the input data (e.g., an upcoming storm, rising crime rates, market volatility), and thereby generate predictions and calculates the probability of specific outcomes. [0155] Still referring to FIG. 4, machine-learning algorithms may include at least a supervised machine-learning process 428. At least a supervised machine-learning process 428, as defined herein, include algorithms that receive a training set relating a number of inputs to a number of outputs, and seek to generate one or more data structures representing and/or instantiating one or more mathematical relations relating inputs to outputs, where each of the one or more mathematical relations is optimal according to some criterion specified to the algorithm using some scoring function. For instance, a supervised learning algorithm may include target as described above as inputs, protection gap as outputs, and a scoring function representing a desired form of relationship to be detected between inputs and outputs; scoring function may, for instance, seek to maximize the probability that a given input and/or combination of elements inputs is associated with a given output to minimize the probability that a given input is not associated with a given output. Scoring function may be expressed as a risk function representing an “expected loss” of an algorithm relating inputs to outputs, where loss is computed as an error function representing a degree to which a prediction generated by the relation is incorrect when compared to a given input-output pair provided in training data 404. Persons skilled in the art, upon reviewing the entirety of this disclosure, will be aware of various possible variations of at least a supervised machine-learning process 428 that may be used to determine relation between inputs and outputs. Supervised machine-learning processes may include classification algorithms as defined above.[156] Updating may be performed, in neural networks, using one or more back-propagation algorithms. Iterative and/or recursive updates to weights, biases, coefficients, or other parameters as described above may be performed until currently available training data is exhausted and/or until a convergence test is passed, where a “convergence test” is a test for a condition selected as indicating that a model and/or weights, biases, coefficients, or other parameters thereof has reached a degree of accuracy. A convergence test may, for instance, compare a difference between two or more successive errors or error function values, where differences below a threshold amount may be taken to indicate convergence. Alternatively or additionally, one or more errors and/or error function values evaluated in training iterations may be compared to a threshold. [156-162] elaborate on the matter)
Regarding claim 11, Arriaga and Zhang teach The method of claim 10, wherein the user preferences indicate, for each test pair of the plurality of test pairs, whether the user accepted or rejected the query written in structured query language programming language corresponding to the test question as an accurate conversion of the test question. (Arriaga [0045] input by a user, and/or previous iterations of processing. A target classifier may be configured to receive as input and categorize components of dataset 120 to one or more target categorizations. In some cases, processor 108 and/or computing device 104 may then select any elements within dataset 120 containing a similar label and/or grouping and group them together. In some cases, dataset 120 may be classified using a classifier machine learning model. In some cases classifier machine learning model may be trained using training data correlating a plurality of datasets 120 correlated to a plurality of target categorizations. In an embodiment, a particular element within dataset 120 may be correlated to a particular target categorization. In some cases, classifying dataset 120 may include classifying dataset 120 as a function of the classifier machine learning model. In some cases classifier training data may be generated through input by a user. In some cases, classifier machine learning model may be trained through user feedback wherein a user may indicate whether a particular element corresponds to a particular class. In some cases, classifier machine learning model may be trained using inputs and outputs based on previous iterations. In some cases, a user may input previous dataset 120 and corresponding target categorizations wherein classifier machine learning model may be trained based on the input.[0046] With continued reference to FIG. 1, in some embodiments, classifier training data may be iteratively updated using feedback. Feedback, in some embodiments, may include user feedback. For example, user feedback may include a rating, such as a rating from 1-10, 1-100, −1 to 1, “happy,” “sad,” and the like. In some embodiments, user feedback may rate a user's satisfaction with the target categorization. In some embodiments, feedback may include outcome data. “Outcome data,” for the purposes of this disclosure, is data including an outcome of a process. As a non-limiting example, outcome data may include information regarding whether a target made a purchase, whether a target has continued communications, and the like. Iteratively updating classifier training data may include removing datasets and target categorizations from classifier training data as a function of negative or unfavorable feedback. In some embodiments, each datasets and target categorization within classifier training data may have an associated weight. That weight may be adjusted based on feedback. For example, the weight may be increased in response to positive or favorable feedback, while the weight may be decreased in response to negative or unfavorable feedback.[0058] With continued reference to FIG. 1, in some embodiments, protection training data 164 may be iteratively updated using feedback. Feedback, in some embodiments, may include user feedback. For example, user feedback may include a rating, such as a rating from 1-10, 1-100, −1 to 1, “happy,” “sad,” and the like. In some embodiments, user feedback may rate a user's satisfaction with the identified protection gap 152. In some embodiments, feedback may include outcome data.[0060] trained through user feedback wherein a user may indicate whether a particular element corresponds to a particular protection categorization 172. In some cases, protection machine learning model 160 may be trained using inputs and outputs based on previous iterations. In some cases, a user may input previous modified datasets 144 and corresponding protection categorizations 172 wherein protection machine learning model 160 may be trained based on the input. [55-60 and 156-163] elaborates on the matter)
Claim 13-17 and 19-21 are rejected under 35 U.S.C. 103 as being unpatentable over US 20250054068 A1; Arriaga; Jason (hereinafter Arriaga) in view of US 20250190449 A1; Zhang; Bo et al. (hereinafter Zhang) and US 20150379427 A1; DIRAC; LEO PARKER et al (hereinafter Dirac).
Regarding claim 13, Arriaga teaches A method comprising: identifying, by a processing system including at least one processor, a user of a text-to-structured query language system; (Arriaga [FIG.1] shows the system identifying a user of a text-to-query written in structured query language programming language system[0004] The memory instructs the processor to generate one or more target profiles as function of the modified dataset, the one or more protection gaps, and a user input. The memory instructs the processor to generate a video report as a function of the one or more target profiles. [0068] natural language processing tools and AI-driven response systems that allow the avatar to interpret user inputs accurately and respond appropriately in real-time. [138-143] elaborate on the matter) presenting, by the processing system to the user a plurality of text pairs, wherein each test pair of the plurality of test pairs comprises: a question phrased in natural language and a machine learning model-generated query written in structured query language programming language corresponding to the question; (Arriaga [0072] The processor may utilize natural language processing (NLP) techniques to understand the context and significance of the text, allowing it to structure a script that logically sequences the information in a way that's easy for users to follow. Once the script is prepared, Processor 108 may segment the script into discrete sections corresponding to different aspects of the insurance information, such as coverage details, risk assessments, and procedural guides. Each segment may then be assigned specific visual and auditory features that will complement and enhance the textual content. For example, segments explaining complex insurance terms might be paired with visual aids like charts or diagrams, [134] In a non-limiting example, the trained predictive model may be validated and tested using a separate dataset to evaluate its accuracy and performance. Continuing the previous non-limiting example, validation and texting of the predictive model may ensure that the model can generalize well to new, unseen data. In a non-limiting example, the predictive model may continuously monitor the incoming data. Continuing, the predictive model may detect potential risks based on the input data (e.g., an upcoming storm, rising crime rates, market volatility), and thereby generate predictions and calculates the probability of specific outcomes. [0155] Still referring to FIG. 4, machine-learning algorithms may include at least a supervised machine-learning process 428. At least a supervised machine-learning process 428, as defined herein, include algorithms that receive a training set relating a number of inputs to a number of outputs, and seek to generate one or more data structures representing and/or instantiating one or more mathematical relations relating inputs to outputs, where each of the one or more mathematical relations is optimal according to some criterion specified to the algorithm using some scoring function. For instance, a supervised learning algorithm may include target as described above as inputs, protection gap as outputs, and a scoring function representing a desired form of relationship to be detected between inputs and outputs; scoring function may, for instance, seek to maximize the probability that a given input and/or combination of elements inputs is associated with a given output to minimize the probability that a given input is not associated with a given output. Scoring function may be expressed as a risk function representing an “expected loss” of an algorithm relating inputs to outputs, where loss is computed as an error function representing a degree to which a prediction generated by the relation is incorrect when compared to a given input-output pair provided in training data 404. Persons skilled in the art, upon reviewing the entirety of this disclosure, will be aware of various possible variations of at least a supervised machine-learning process 428 that may be used to determine relation between inputs and outputs. Supervised machine-learning processes may include classification algorithms as defined above.[156] Updating may be performed, in neural networks, using one or more back-propagation algorithms. Iterative and/or recursive updates to weights, biases, coefficients, or other parameters as described above may be performed until currently available training data is exhausted and/or until a convergence test is passed, where a “convergence test” is a test for a condition selected as indicating that a model and/or weights, biases, coefficients, or other parameters thereof has reached a degree of accuracy. A convergence test may, for instance, compare a difference between two or more successive errors or error function values, where differences below a threshold amount may be taken to indicate convergence. Alternatively or additionally, one or more errors and/or error function values evaluated in training iterations may be compared to a threshold. [156-162] elaborate on the matter) recording, by the processing system for each test pair of the plurality of test pairs in a signature table for the user… (Arriaga [0045] input by a user, and/or previous iterations of processing. A target classifier may be configured to receive as input and categorize components of dataset 120 to one or more target categorizations. In some cases, processor 108 and/or computing device 104 may then select any elements within dataset 120 containing a similar label and/or grouping and group them together. In some cases, dataset 120 may be classified using a classifier machine learning model. In some cases classifier machine learning model may be trained using training data correlating a plurality of datasets 120 correlated to a plurality of target categorizations. In an embodiment, a particular element within dataset 120 may be correlated to a particular target categorization. In some cases, classifying dataset 120 may include classifying dataset 120 as a function of the classifier machine learning model. In some cases classifier training data may be generated through input by a user. In some cases, classifier machine learning model may be trained through user feedback wherein a user may indicate whether a particular element corresponds to a particular class. In some cases, classifier machine learning model may be trained using inputs and outputs based on previous iterations. In some cases, a user may input previous dataset 120 and corresponding target categorizations wherein classifier machine learning model may be trained based on the input.[0046] With continued reference to FIG. 1, in some embodiments, classifier training data may be iteratively updated using feedback. Feedback, in some embodiments, may include user feedback. For example, user feedback may include a rating, such as a rating from 1-10, 1-100, −1 to 1, “happy,” “sad,” and the like. In some embodiments, user feedback may rate a user's satisfaction with the target categorization. In some embodiments, feedback may include outcome data. “Outcome data,” for the purposes of this disclosure, is data including an outcome of a process. As a non-limiting example, outcome data may include information regarding whether a target made a purchase, whether a target has continued communications, and the like. Iteratively updating classifier training data may include removing datasets and target categorizations from classifier training data as a function of negative or unfavorable feedback. In some embodiments, each datasets and target categorization within classifier training data may have an associated weight. That weight may be adjusted based on feedback. For example, the weight may be increased in response to positive or favorable feedback, while the weight may be decreased in response to negative or unfavorable feedback.[0058] With continued reference to FIG. 1, in some embodiments, protection training data 164 may be iteratively updated using feedback. Feedback, in some embodiments, may include user feedback. For example, user feedback may include a rating, such as a rating from 1-10, 1-100, −1 to 1, “happy,” “sad,” and the like. In some embodiments, user feedback may rate a user's satisfaction with the identified protection gap 152. In some embodiments, feedback may include outcome data.[0060] trained through user feedback wherein a user may indicate whether a particular element corresponds to a particular protection categorization 172. In some cases, protection machine learning model 160 may be trained using inputs and outputs based on previous iterations. In some cases, a user may input previous modified datasets 144 and corresponding protection categorizations 172 wherein protection machine learning model 160 may be trained based on the input. [55-60 and 156-163] elaborates on the matter) to predict likelihood that of target features should be present in a new query written in structured query language programming language based on a set of query features is determined to be present... (Arriaga [0053] include dataset 120, target data .... inference engine may determine that a particular rule relating to policy limits should be selected when the elements within modified dataset 144 discuss a particular policy limit. Similarly, a particular rule relating to types of protection may be selected when elements within modified dataset 144 indicate types of protection. In some cases, gap finder module 156 may receive elements within modified dataset 144 and/or target data 124 and make calculations ... within target data 124. For example, web crawler may be configured to retrieve an estimate of the target's property using estimates from one or more property websites. [0101] Still referring to FIG. 1, the target machine learning model may include a large language model (LLM). A “large language model,” as used herein, is a deep learning data structure that can recognize, summarize, translate, predict and/or generate text and other content based on knowledge gained from massive datasets. Large language models may be trained on large sets of data. Training sets may be drawn from diverse sets of data such as, as non-limiting examples, novels, blog posts, articles, emails, unstructured data, electronic records, and the like. In some embodiments, training sets may include a variety of subject matters, such as, as nonlimiting examples, weather reports, insurance policies, insurance claims, property damage reports, emails, user communications, advertising documents, newspaper articles, and the like. In some embodiments, training sets of an LLM may include information from one or more public or private databases. As a non-limiting example, training sets may include databases associated with an entity. In some embodiments, training sets may include portions of documents associated with the electronic records 112 correlated to examples of outputs. In an embodiment, an LLM may include one or more architectures based on capability requirements of an LLM. Exemplary architectures may include, without limitation, GPT (Generative Pretrained Transformer), BERT (Bidirectional Encoder Representations from Transformers), T5 (Text-To-Text Transfer Transformer), and the like. Architecture choice may depend on a needed capability such generative, contextual, or other specific capabilities.) [102] A model such as an LLM may learn by adjusting its parameters during the training process to minimize a defined loss function, which measures the difference between predicted outputs and ground truth. Once a model has been generally trained, the model may then be specifically trained to fine-tune the pretrained model on task-specific data to adapt it to the target task. Fine-tuning may involve training a model with task-specific training data, adjusting the model's weights [107,116 and 151] elaborate on the matter) Arriaga lacks explicitly and orderly teaching training, by the processing system, a machine learning model, using the signature table as training data… query features is determined to be present in the new query written in structured query language programming language. However Zhang teaches training, by the processing system, a machine learning model, using the signature table as training data… query features is determined to be present in the new query written in structured query language programming language. (Zhang [0005] analyze the structured data or to be trained in the use of a particular application interface. The system may also be configured to perform continuous learning such that user feedback and preferences may be provided based on the response, and the system may adjust parameters of the prompt processing and agent selection processes to increase accuracy and user-satisfaction with the generative AI-assisted analytics performed by the system. [0031] The structured data set 154 may include multiple file types or a single file type, and may include or be organized according to multiple schemas, multiple data table formats, multiple data types, or a combination thereof. [0040]The generative AI agents 126 include any types of generative AI agents that are capable of performing analytics or mathematical operations on input data, such as a comma-separated value (CSV) agent, a structured query language (SQL) ... [0040] AI models (referred to herein collectively as “the AI models 125”) that are trained to perform the tasks described with reference to the agent orchestrator 124. For example, the AI models 125 may include a trained AI classifier or ML classifier that is configured to select one or more of the generative AI agents 126 based on the prompt 170 and the structured data 172. [0044] user feedback 180 that indicates the user's feedback and any preferences associated with the prompt 170 and the response 110, the GUI 174, and/or the visual elements 176. For example, the user feedback 180 may indicate an accuracy score and a preference for a line graph that is related to a user prompt that requested performance of some statistical analysis and “a graph displaying the results.” The computing device 102 may modify the knowledge graph 120 based on the prompt 170, the response 110, and the user feedback 180. For example, one or more nodes may be added to the knowledge graph 120 based on preferences indicated by the user feedback [0056] sub-intents related to user prompts and responses in order to facilitate adaptation to user feedback of responses output by the generative AI analytics engine 200. In such an example, a received user feedback that indicates a user's feedback and preferences related to a response may be used to update weights and relationships in the knowledge graph in real-time [90-96] elaborate on the matter [FIG.1 in conjunction with FIG.3] show actions and processes based on an input query written in structured query language programming language based on a presence of a query feature in the input query written in structured query language programming language; wherein the machine learning model has been trained using a signature table that captures user preferences with respect to a plurality of test pairs) Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to take all prior methods and make the addition of Zhang in order to help efficiently improve the querying output of the system (Zhang [0003] Additionally, these analytical applications are typically designed to with the functionality as the primary concern, often without prioritizing ease of use, particularly by customers who are inexperienced in using data-driven user interfaces. As technology has advanced, artificial intelligence and machine learning has been leveraged to create generative artificial intelligence models, such as large language models (LLMs), that can create novel text outputs that can be used to improve ease of use and user interactivity with user interfaces of applications. However, LLMs are typically not suited for performing analytics and can be prone to “hallucinations,” such as generating incorrect information in response to a question for which the LLM is unable to correctly answer. As such, although LLMs and generative artificial intelligence have improved usability of some applications, these improvements have not extended to analytical applications and to structured data set contexts.[0004] prompt optimization operations, or a combination thereof, to modify the prompt such that the prompt is more useful to the agent orchestrator (e.g., so that performance of the agent orchestrator is improved as compared to using the unmodified user prompt). The agent orchestrator may include a trained AI classifier configured to select the one or more generative AI agents from multiple available generative AI agents that are configured to perform analytics tasks on numerical and structured data. As non-limiting examples, the generative AI agents may include a CSV agent, an SQL agent, a Spark agent, a semantic search agent, or the like. The agent orchestrator may select the generative AI agents based on agent features associated with the available generative AI agents, intent features associated with the intent determined from the user prompt, data schema features associated with the structured data set, or a combination thereof, and optionally based on a greedy parameter to enable exploration of agents with lower scores based on current knowledge but that may have the potential to yield better results.[0005] a table (or portion thereof), or other visual elements in addition to text output and/or numerical output to provide a user with information to enable efficient derivation of insights and decision-making capabilities. Such an output may have greater utility than a wholly text or numerical output, thereby improving a user experience and enabling easier and more accurate decision-making and understanding of insights unlocked from structured data without requiring the user to manually analyze the structured data or to be trained in the use of a particular application interface. The system may also be configured to perform continuous learning such that user feedback and preferences may be provided based on the response, and the system may adjust parameters of the prompt processing and agent selection processes to increase accuracy and user-satisfaction with the generative AI-assisted analytics performed by the system.[0031] the prompt 170 and/or the structured data set 154 and to use the agent orchestrator 124 to select one or more of the generative AI agents 126 for providing an accurate and efficient response to the prompt 170, as further described herein [0043] answer the question(s) included in the prompt 170 to the user of the user device 150, to enable quick and efficient understanding by the user, thereby enabling fast and effective decision making based on insights drawn from the structured data set [FIG.1] shows overall visual) The combination lack and explicitly teaching recording, by the processing system for each test pair of the plurality of test pairs in a signature table for the user whether the user accepted or rejected the each test pair; However Dirac teaches recording, by the processing system for each test pair of the plurality of test pairs in a signature table for the user whether the user accepted or rejected the each test pair; (Dirac [0160] If the accuracy/quality measures 2630 are satisfactory, the candidate model 2620 may be designated as an approved model 2640 in the depicted embodiment. Otherwise, any of several techniques may be employed in an attempt to improve the quality or accuracy of the model's predictions. Model tuning 2672 may comprise modifying the set of independent variables being used for the predictions, changing model execution parameters (such as a minimum bucket size or a maximum tree depth for tree-based classification models), and so on, and executing additional training runs 2618. Model tuning may be performed iteratively using the same training and test sets, varying some combination of independent variables and parameters in each iteration in an attempt to enhance the accuracy or quality of the results. In another approach to model improvement, changes 2674 may be made to the training and test data sets for successive training-and-evaluation iterations. For example, the input data set may be shuffled (e.g., at the chunk level and/or at the observation record level), and a new pair of training/test sets may be obtained for the next round of training. In another approach, the quality of the data may be improved by, for example, identifying observation records whose independent variable values appear to be invalid or outliers, and deleting such observation records from the data set. [161] observation records 2702 is split five different ways to obtain respective training sets 2720 (e.g., 2720A-2720E) each comprising 80% of the data, and corresponding test sets 2710 (e.g., 2710A-2710E) comprising the remaining 20% of the data. Each of the training sets 2720 may be used to train a model, and the corresponding test set 2710 may then be used to evaluate the model. For example, in cross-validation iteration 2740A, the model may be trained using training set 2720A and then evaluated using test set 2710A. Similarly, in cross-validation iteration 2740B, a different training set 2720B (shown in two parts, part 1 and part 2 in FIG. 27) comprising 80% of the input data may be used, and a different test set 2710B may be used for evaluating the model. The cross-validation example illustrated in FIG. 27 may be referred to as “5-fold cross validation” (because of the number of different training/test set pairs generated and the corresponding number of training-and-evaluation iterations.) The MLS may implement an API allowing a client to request k-fold cross validation in some embodiments, where k is an API parameter indicating the number of distinct training sets (and corresponding test sets) to be generated for training a specified model using the same underlying input data set.[0231] After an FP proposal is approved by a client, it may be used for subsequent executions of the model (i.e., processed variables produced using the FP proposal may be used as input variables used to train the model and to make predictions using the model), potentially for many different production-mode data sets. A given client may submit several different model creation requests to the service, approve respective FP proposals for each model, and then utilize the approved models for a while. In some implementations, clients may wish to view the success rate with respect to their prediction run-time goals for various models after they are approved. FIG. 49 is an example of a programmatic dashboard interface that may enable clients to view the status of a variety of machine learning model runs ... [0528] generate, based at least in part on the quality estimate and at least in part on the cost estimate, a feature processing proposal to be provided to the client for approval, wherein the feature processing proposal includes a recommendation to implement the particular feature processing transformation; and [0529] in response to an indication of approval from the client, execute a machine learning model trained using a particular processed variable obtained from the particular feature processing transformation. [FIG.1] shows corresponding system which involve the visual flow of wherein recording that the user accepted a subset of the plurality of test pairs...) Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to take all prior methods and make the addition of Dirac's feature processing and integration methods in order to improve the systems output via improved predictions (Dirac [0003] Traditionally, expertise in statistics and in artificial intelligence has been a prerequisite for developing and using machine learning models. For many business analysts and even for highly qualified subject matter experts, the difficulty of acquiring such expertise is sometimes too high a barrier to be able to take full advantage of the large amounts of data potentially available to make improved business predictions and decisions. [0004] The quality of the results obtained from machine learning algorithms may depend on how well the empirical data used for training the models captures key relationships among different variables represented in the data, and on how effectively and efficiently these relationships can be identified. Depending on the nature of the problem that is to be solved using machine learning, very large data sets may have to be analyzed in order to be able to make accurate predictions [0032] FIG. 26 illustrates an example of an iterative procedure that may be used to improve the quality of predictions made by a machine learning model, according to at least some embodiments.)
Regarding claim 14, Arriaga, Zhang and Dirac teach The method of claim 13, wherein the plurality of test pairs is generated using a retrieval-augmented generation pipeline. (Arriga[0096] With continued reference to FIG. 1, the LLM may include a Retrieval-Augmented Generation (RAG) system. As used in the current disclosure, a “retrieval-Augmented Generation system” is a hybrid artificial intelligence technique that enhances the performance of generative models by integrating them with a retrieval component. In this approach, the system may first retrieve relevant information from a database or a large corpus of documents, which is then used to inform and guide the generation process of the language model. This method may be particularly useful in scenarios where accuracy and factuality are crucial, such as content creation, question answering, and decision support systems. By grounding the responses of the generative model in actual data, RAG helps mitigate the issue of generating misleading or incorrect information, thus improving the reliability and relevance of the outputs produced by the model. RAG may enhance the capabilities of language models by combining the generative power of models like large language models (LLMs) with the ability to retrieve factual information from external databases. In the context of an insurance-related application, RAG allows [0155] Still referring to FIG. 4, machine-learning algorithms may include at least a supervised machine-learning process 428. At least a supervised machine-learning process 428, as defined herein, include algorithms that receive a training set relating a number of inputs to a number of outputs, and seek to generate one or more data structures representing and/or instantiating one or more mathematical relations relating inputs to outputs, where each of the one or more mathematical relations is optimal according to some criterion specified to the algorithm using some scoring function. For instance, a supervised learning algorithm may include target as described above as inputs, protection gap as outputs, and a scoring function representing a desired form of relationship to be detected between inputs and outputs; scoring function may, for instance, seek to maximize the probability that a given input and/or combination of elements inputs is associated with a given output to minimize the probability that a given input is not associated with a given output. Scoring function may be expressed as a risk function representing an “expected loss” of an algorithm relating inputs to outputs, where loss is computed as an error function representing a degree to which a prediction generated by the relation is incorrect when compared to a given input-output pair provided in training data 404. Persons skilled in the art, upon reviewing the entirety of this disclosure, will be aware of various possible variations of at least a supervised machine-learning process 428 that may be used to determine relation between inputs and outputs. Supervised machine-learning processes may include classification algorithms as defined above. [165] using pipelining, parallel processing, or the like; such a hardware unit may be optimized for such processes by, for instance, including dedicated circuitry for matrix and/or signal processing operations that includes [FIG.1] overall visual)
Regarding claim 15, Arriaga, Zhang and Dirac teach The method of claim 13, wherein the signature table further records, for each test pair: a set of query features, a set of target features, and metadata. (Arriaga [0045] With continued reference to FIG. 1, a “classifier,” as used in this disclosure is a machine-learning model, such as a mathematical model, neural net, or program generated by a machine learning algorithm known as a “classification algorithm,” as described in further detail below, that sorts inputs into categories or bins of data, outputting the categories or bins of data and/or labels associated therewith. Classifiers as described throughout this disclosure may be configured to output at least a datum that labels or otherwise identifies a set of data that are clustered together, found to be close under a distance metric as described below, or the like. In some cases, processor 108 may generate and train a target classifier ... In some cases classifier training data may be generated through input by a user. In some cases, classifier machine learning model may be trained through user feedback wherein a user may indicate whether a particular element corresponds to a particular class. In some cases, classifier machine learning model may be trained using inputs and outputs based on previous iterations. In some cases, a user may input previous dataset 120 and corresponding target categorizations wherein classifier machine learning model may be trained based on the input. [0055] machine-learning model to determine its own outputs for inputs. Training data may contain correlations that a machine-learning process may use to model relationships between two or more categories of data elements. Exemplary inputs and outputs may come from a database, such as any database described in this disclosure, or be provided by a user. In other embodiments, a machine-learning module may obtain a training set by querying a communicatively connected database that includes past inputs and outputs. Training data may include inputs from various types of databases, resources, and/or user inputs and outputs correlated to each of those inputs so that a machine-learning model may determine an output. Correlations may indicate causative and/or predictive links between data, which may be modeled as relationships, such as mathematical relationships, by machine-learning models, as described in further detail below. In one or more embodiments, training data may be formatted and/or organized by categories of data elements by, for example, associating data elements with one or more descriptors corresponding to categories of data elements. As a non-limiting example, training data may include data entered in standardized forms by persons or processes, such that entry of a given data element in a given field in a form may be mapped to one or more descriptors of categories. Elements in training data may be linked to categories by tags, tokens, or other data elements [102] one or more machine learning models may include setting the parameters of the one or more models (weights and biases) either randomly or using a pretrained model. Generally training one or more machine learning models on a large corpus of text data can provide a starting point for fine-tuning on a specific task. A model such as an LLM may learn by adjusting its parameters during the training process to minimize a defined loss function, which measures the difference between predicted outputs and ground truth. Once a model has been generally trained, the model may then be specifically trained to fine-tune the pretrained model on task-specific data to adapt it to the target task. Fine-tuning may involve training a model with task-specific training data [133-137] elaborate on the matter)
Regarding claim 16, Arriaga, Zhang and Dirac teach The method of claim 15, wherein the set of query features is used to determine which features of the set of target features should be applied to the machine learning model-generated query written in structured query language programming language. (Arriaga [0061] target data 124 and/or modified dataset 144 is associated to a vehicle protection categorization 172. In one or more embodiments, the chatbot may be configured to define terms for a user and/or target. For example, a target may input a query into chatbot asking what a particular term means wherein chatbot may be configured to define the terms. In one or more embodiments, the chatbot may provide visual aid, tool tips, and videos in order to define complex terms and help an individual understand one or more terms. In one or more embodiments, chatbot may utilize a machine learning model, such as any machine learning model as described in this disclosure wherein training data may be used to generate outputs that can educate individuals on complex terms. In one or more embodiments, training data may include previous inputs by previous individuals into the chatbot and outputs that provide visual aid, videos, and the like. In an embodiment, each input by a user into chatbot may be used to train the machine learning model wherein responses such as “I still don't understand” by the individual may indicate that a particular term requires more a more precise definition. In one or more embodiments, user input into chatbot system may be used to train the machine learning model wherein chatbot may be configured to provide more accurate results on every iteration. In one or more embodiments, training of the machine learning model, and as a result training chatbot, may allow for quicker and more efficient communication between chatbot and an individual. In an embodiment, training data may allow for shorter communications with chatbot, and as result, less strain on one or more computing devices configured to generate chatbot. [0102] adjusting the model's weights to optimize performance for the particular task. In some cases, this may include optimizing the model's performance by fine-tuning hyperparameters such as learning rate, batch size, and regularization. Hyperparameter tuning may help in achieving the best performance and convergence during training. In an embodiment, fine-tuning a pretrained model such as an LLM may include fine-tuning the pretrained model using Low-Rank Adaptation (LoRA). As used in this disclosure, “Low-Rank Adaptation” is a training technique for large language models that modifies a subset of parameters in the model. Low-Rank Adaptation may be configured to make the training process more computationally efficient by avoiding a need to train an entire model from scratch. In an exemplary embodiment, a subset of parameters that are updated may include parameters that are associated with a specific task or domain. [133] classifying target profile 176 into cohorts using a classification model. For example, without limitation, target profile 176 may be classified into cohorts such as “young professionals,” “families,” retirees,” and the like. In another non-limiting example the classification model may analyze user preferences, behaviors, interest, interactions with the platform, and the like to determine the cohorts and assign target profile 176 to a specific cohort. In a non-limiting example, the classification model may include one or more algorithms and/or machine learning models to analyze and classify target profiles 176. In a non-limiting example, the personalized content feed may be integrated into the unified dashboard. Without limitation, the personalized content feed may interactively and continuously learn and adjust recommendations based on the latest user behaviors, interactions, interests, and the like to ensure that the personalized content feed remains relevant and useful to the user. In some embodiments, training data for the classification model may include exemplary target profiles correlated to cohort labels. [134-138] elaborate on the matter)
Regarding claim 17, Arriaga, Zhang and Dirac teach The method of claim 15, wherein the set of query features includes at least one of: a table referenced by the new query written in structured query language programming language, a number of subqueries contained in the new query written in structured query language programming language, a number of joins contained in the new query written in structured query language programming language, a number of where clauses contained in the new query written in structured query language programming language, a number of fields of the table returned by the new query written in structured query language programming language, or whether a top-level new query written in structured query language programming language performs aggregation. (Arriaga [0108] With continued reference to FIG. 1, multi-headed attention in encoder may apply a specific attention mechanism called self-attention. Self-attention allows models such as an LLM or components thereof to associate each word in the input, to other words. As a non-limiting example, an LLM may learn to associate the word “you”, with “how” and “are”. It's also possible that an LLM learns that words structured in this pattern are typically a question and to respond appropriately. In some embodiments, to achieve self-attention, input may be fed into three distinct fully connected neural network layers to create query, key, and value vectors. A query vector may include an entity's learned representation for comparison to determine attention score. A key vector may include an entity's learned representation for determining the entity's relevance and attention weight. A value vector may include data used to generate output representations. Query, key, and value vectors may be fed through a linear layer; then, the query and key vectors may be multiplied using dot product matrix multiplication in order to produce a score matrix. The score matrix may determine the amount of focus for a word should be put on other words (thus, each word may be a score that corresponds to other words in the time-step). [0109] Still referencing FIG. 1, in order to use self-attention in a multi-headed attention computation, query, key, and value may be split into N vectors before applying self-attention. Each self-attention process may be called a “head.” Each head may produce an output vector and each output vector from each head may be concatenated into a single vector. This single vector may then be fed through the final linear layer discussed above. In theory, each head can learn something different from the input, therefore giving the encoder model more representation power. [138] the API query may include authentication information such as API keys, tokens, or other credentials required to authenticate the request with the third-party application. In another non-limiting example, the API query may include the type of request being made, such as “GET,” “POST,” “PUT,” and the like, depending on the action required. For example, without limitation the API may query to an insurance carrier's APU to retrieve updated policy information. [56-61] further elaborate on the query features)
Regarding claim 19, Arriaga, Zhang and Dirac teach The method of claim 13, wherein the machine learning model is at least one of: a deep learning model or a boosted tree. (Arriaga [0045] With continued reference to FIG. 1, a “classifier,” as used in this disclosure is a machine-learning model, such as a mathematical model, neural net, or program generated by a machine learning algorithm known as a “classification algorithm,” as described in further detail below, that sorts inputs into categories or bins of data, outputting the categories or bins of data and/or labels associated therewith. Classifiers as described throughout this disclosure may be configured to output at least a datum that labels or otherwise identifies a set of data that are clustered together, found to be close under a distance metric as described below, or the like. In some cases, processor 108 may generate and train a target classifier configured to receive dataset 120 and output one or more target categorizations. Processor 108 and/or another device may generate a classifier using a classification algorithm, defined as a process whereby a computing device 104 derives a classifier from training data. In some cases target classifier may use data to prioritize the order of labels within dataset 120. Classification may be performed using, without limitation, linear classifiers such as without limitation logistic regression and/or naive Bayes classifiers, nearest neighbor classifiers such as k-nearest neighbors classifiers, support vector machines, least squares support vector machines, fisher's linear discriminant, quadratic classifiers, decision trees, boosted trees, random forest classifiers, learning vector quantization, and/or neural network-based classifiers. A target classifier may be trained with training data correlating dataset 120 to descriptor groupings such as simplifiers, multipliers, and the like. [101] multi-model neural network and combination thereof that may be implemented by apparatus 100 in consistent with this disclosure. Still referring to FIG. 1, the target machine learning model may include a large language model (LLM). A “large language model,” as used herein, is a deep learning data structure that can recognize, summarize, translate, predict and/or generate text and other content based on knowledge gained from massive datasets. Large language models may be trained on large sets of data. Training sets may be drawn from diverse sets of data such as, as non-limiting examples, novels, blog posts, articles, emails, unstructured data, electronic records, and the like. In some embodiments, training sets may include a variety of subject matters, such as, as nonlimiting examples, weather reports, insurance policies, insurance claims, property damage reports, emails, user communications, advertising documents, newspaper articles, and the like. In some embodiments, training sets of an LLM may include information from one or more public or private databases. As a non-limiting example, training sets may include databases associated with an entity. In some embodiments, training sets may include portions of documents associated with the electronic records 112 correlated to examples of outputs. In an embodiment, an LLM may include one or more architectures based on capability requirements of an LLM. Exemplary architectures may include, without limitation, GPT (Generative Pretrained Transformer), BERT (Bidirectional Encoder Representations from Transformers), T5 (Text-To-Text Transfer Transformer), and the like. Architecture choice may depend on a needed capability such generative, contextual, or other specific capabilities. [160] Machine-learning algorithms may include naïve Bayes methods. Machine-learning algorithms may include algorithms based on decision trees, such as decision tree classification or regression algorithms. Machine-learning algorithms may include ensemble methods such as bagging meta-estimator, forest of randomized trees, AdaBoost, gradient tree boosting, and/or voting classifier methods. Machine-learning algorithms may include neural net algorithms, including convolutional neural net processes.)
Regarding claim 20, Arriaga, Zhang and Dirac teach The method of claim 13, wherein the machine learning model is one of a plurality of machine learning models that work together, wherein each machine learning model of the plurality of machine learning models is trained to predict the presence of a different target feature in a given query written in structured query language programming language. (Arriaga [0055] With continued reference to FIG. 1, gap finder module 156 may include a protection machine learning model 160. Processor 108 and/or gap finder module 156 may use a machine learning module, such as a protection machine learning module for the purposes of this disclosure, to implement one or more algorithms or generate one or more machine-learning models, such as a protection machine learning model 160, to calculate at least one protection gap 152. However, the machine learning module is exemplary and may not be necessary to generate one or more machine learning models and perform any machine learning described herein. In one or more embodiments, one or more machine-learning models may be generated using training data. Training data may include inputs and corresponding predetermined outputs so that a machine-learning model may use correlations between the provided exemplary inputs and outputs to develop an algorithm and/or relationship that then allows machine-learning model to determine its own outputs for inputs. Training data may contain correlations that a machine-learning process may use to model relationships between two or more categories of data elements. [0056] With continued reference to FIG. 1, in one or more embodiments, a machine-learning module may be generated using training data. Training data may include inputs and corresponding predetermined outputs so that machine-learning module may use the correlations between the provided exemplary inputs and outputs to develop an algorithm and/or relationship that then allows machine-learning module to determine its own outputs for inputs. Training data may contain correlations that a machine-learning process may use to model relationships between two or more categories of data elements. The exemplary inputs and outputs may come from a database, such as any database described in this disclosure, or be provided by a user such as a prospective employee, and/or an employer and the like. In other embodiments, machine-learning module may obtain a training set by querying a communicatively connected database that includes past inputs and outputs. Training data may include inputs from various types of databases, resources, and/or user inputs and outputs correlated to each of those inputs so that a machine-learning module may determine an output. Correlations may indicate causative and/or predictive links between data, which may be modeled as relationships, such as mathematical relationships, by machine-learning processes[0088] Still referring to FIG. 1, in some cases, generative machine learning models may include one or more generative models. As described herein, “generative models” refers to statistical models of the joint probability distribution P(X,Y) on a given observable variable x, representing features or data that can be directly measured or observed (e.g., target profiles 176) and target variable y, representing the outcomes or labels that one or more generative models aims to predict or generate (e.g., video report 178). In some cases, generative models may rely on Bayes theorem to find joint probability; for instance, and without limitation, Naïve Bayes classifiers may be employed by processor 108 to categorize input data such as, without limitation, target profiles 176 into different stewardship files [101-102] goes into detail on the plurality of ML models and corresponding feature prediction method.)
Regarding claim 21, Arriaga and Zhang teach The method of claim 1, the combination lack explicitly and orderly teaching wherein the user preferences indicate that the user accepted a subset of the plurality of test pairs in which a first query feature and a first target feature were both present, and wherein the recommendation comprises a recommendation to add the first target feature to the query written in structured query language programming language in response to the first query feature being detected in the query written in structured query language programming language. However Dirac teaches wherein the user preferences indicate that the user accepted a subset of the plurality of test pairs in which a first query feature and a first target feature were both present, and wherein the recommendation comprises a recommendation to add the first target feature to the query written in structured query language programming language in response to the first query feature being detected in the query written in structured query language programming language. (Dirac [161] different test set 2710B may be used for evaluating the model. The cross-validation example illustrated in FIG. 27 may be referred to as “5-fold cross validation” (because of the number of different training/test set pairs generated and the corresponding number of training-and-evaluation iterations.) The MLS may implement an API allowing a client to request k-fold cross validation in some embodiments, where k is an API parameter indicating the number of distinct training sets (and corresponding test sets) to be generated for training a specified model using the same underlying input data set [0231] After an FP proposal is approved by a client, it may be used for subsequent executions of the model (i.e., processed variables produced using the FP proposal may be used as input variables used to train the model and to make predictions using the model), potentially for many different production-mode data sets. A given client may submit several different model creation requests to the service, approve respective FP proposals for each model, and then utilize the approved models for a while. In some implementations, clients may wish to view the success rate with respect to their prediction run-time goals for various models after they are approved. FIG. 49 is an example of a programmatic dashboard interface that may enable clients to view the status of a variety of machine learning model runs ... [0525] determine, via one or more programmatic interactions with a client of a machine learning service of a provider network, (a) one or more target variables to be predicted using a specified training data set, (b) one or more prediction quality metrics including a particular prediction quality metric, and (c) one or more prediction run-time goals including a particular prediction run-time goal; [0526] identify a set of candidate feature processing transformations to derive a first set of processed variables from one or more input variables of the specified data set, wherein at least a subset of the first set of processed variables is usable to train a machine learning model to predict the one or more target variables, and wherein the set of candidate feature processing transformations includes a particular feature processing transformation [0528] generate, based at least in part on the quality estimate and at least in part on the cost estimate, a feature processing proposal to be provided to the client for approval, wherein the feature processing proposal includes a recommendation to implement the particular feature processing transformation; and [0529] in response to an indication of approval from the client, execute a machine learning model trained using a particular processed variable obtained from the particular feature processing transformation. [FIG.1] shows corresponding system which involve the visual flow of wherein the user preferences indicate that the user accepted a subset of the plurality of test pairs in which a first query feature and a first target feature were both present, and wherein the recommendation comprises a recommendation to add the first target feature to the query written in structured query language programming language in response to the first query feature being detected in the query written in structured query language programming language.) Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to take all prior methods and make the addition of Dirac's feature processing and integration methods in order to improve the systems output via improved predictions (Dirac [0003] Traditionally, expertise in statistics and in artificial intelligence has been a prerequisite for developing and using machine learning models. For many business analysts and even for highly qualified subject matter experts, the difficulty of acquiring such expertise is sometimes too high a barrier to be able to take full advantage of the large amounts of data potentially available to make improved business predictions and decisions. [0004] The quality of the results obtained from machine learning algorithms may depend on how well the empirical data used for training the models captures key relationships among different variables represented in the data, and on how effectively and efficiently these relationships can be identified. Depending on the nature of the problem that is to be solved using machine learning, very large data sets may have to be analyzed in order to be able to make accurate predictions[0032] FIG. 26 illustrates an example of an iterative procedure that may be used to improve the quality of predictions made by a machine learning model, according to at least some embodiments.)
Claim 18 are rejected under 35 U.S.C. 103 as being unpatentable over Arriaga in view of Zhang, Dirac and US 20200409943 A1; DIEFENBACH; Dennis (hereinafter Dennis.)
Regarding Claim 18, Arriaga, Zhang, and Dirac teach The method of claim 15, wherein the set of target features includes at least one of: Arriaga lacks explicitly and orderly teaching whether a top-level query written in structured query language programming language uses select distinct or whether the query written in structured query language programming language uses an is not null predicate. However Dennis teaches whether a top-level query written in structured query language programming language uses select distinct or whether the query written in structured query language programming language uses an is not null predicate. (Dennis [0127] 4. SELECT DISTINCT ?x WHERE {?x dbo:hometown dbr:Saint-Etienne .} [0142] Ability to bridge over implicit relations: the present method allows to bridge over implicit relations. For example, given the question “Give me German mathematicians” the following top-ranked query is computed: SELECT DISTINCT ?x WHERE {?x ?p1 dbr:Mathematician . ?x ?p2 dbr:Germany .}. Here ?p1 is dbo:field, dbo:occupation, dbo:profession and ?p2 is dbo:nationality, dbo:birthPlace, dbo:death Place, dbo:residence. Note that all these properties could be intended for the given question.[FIG.2] shows overall visual of the system ) Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to take all prior methods and make the addition of Dennis in order to create an improved methods of efficiently answering a users question (Dennis [0014] For instance, patent application EP3179384 proposes to perform uncertain interference on each proposition so as to calculate a “confidence” of each proposition and reducing ambiguity.[0017] Consequently, there is still a need for an improved method for parsing a question in a knowledge base which could be multilingual, and/or KB-agnostic.[0059] Indeed, even if the large majority of these queries will be absurd and not related in any way to the user's intent, one of these queries will be necessary the perfect one. And doing so is paradoxically more efficient as it does not require either complex processing or understanding of the language.[0121] Additional criteria are advantageously used, in particular one or more of: [0135] In a more general way, if there is more than one query intended to be executed (if at least one of the next ranked possible queries of a set is to be executed, of if there are several sets separately ranked for different KBs), the confidence score could be calculated for each of these queries intended to be executed, and only the ones with a confidence score above the threshold are effectively executed. The user thus would have to rephrase the question only if none of the queries could be effectively executed.[0144] Permanent system refinement: It is possible to improve the system over time. The system generates multiple queries. This fact can be used to easily create new training dataset. Using these data sets one can refine the ranker to perform better on the asked questions.[0146] The present method could be used in efficient search engines, recommendation engines (for providing suggestions of movies, music, news, books, research articles, goods, etc. to the user), expert systems, etc.)
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to ARYAN D TOUGHIRY whose telephone number is (571)272-5212. The examiner can normally be reached Monday - Friday, 9 am - 5 pm.
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, Aleksandr Kerzhner can be reached at (571) 270-1760. 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.
/ARYAN D TOUGHIRY/Examiner, Art Unit 2165
Prosecution Insights