DETAILED ACTION
This action is responsive to the Application filed 9/09/2024.
Accordingly, claims 1-20 are submitted for prosecution on merits.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claim 1 is rejected under 35 U.S.C. 101 because the claimed invention is directed to a judicial exception (i.e., a law of nature, a natural phenomenon, or an abstract idea) without significantly more. Claim(s) 1 is/are directed to Abstract Idea. The claim(s) does/do not include additional elements that are sufficient to amount to significantly more than the judicial exception because of the 2-step analysis as follows.
Step I
This claim is directed to a method/process category.
Step 2A
Prong one:
The elements recited as: "obtaining... input data associated with at least one user, wherein the input data comprises... natural language input data"; “analyzing.., Markov discloses input data using each of a prompt model and a large language model... wherein the prompt model interprets input data for generating requirements... wherein the large language model processes the requirements for designing... software solutions" and "generating... software solution information... based on the analyzing" are limitations that recite abstract concepts falling into the Abstract Idea groupings of (i) mental processes and (ii) methods of organizing human activity. MPEP 2106.04(a)(2). In regard to (i), the steps of “obtaining” and “analyzing” (as recited) are conceptually viewed as gathering natural language inputs, interpreting user requirements, organizing information to design software solutions, and outputting recommendations– e.g. generating a solution – are operations that can be performed in the human mind (or by a human using pen and paper). That is, a human system designer can receive a client request (natural language input), interpret the requirements, apply general design rules (acting as the prompt/LLM abstraction), formulate a software solution, and transmit/store those recommendations. MPEP 2106.04(a)(2)(III)
In regard to (ii), designing software solutions based on user requirements is an activity that represents commercial/administrative interactions and basic operational planning. MPEP 2106.04(a)(2)(II)
Prong two:
The elements recited as “processing device”, “communication device”, “storage device” are disassociated requirements with respect to operation of the Abstract Idea. These are perceived as generic components or hardware construed in a high level of generality and can be viewed as mere conduit to execute the abstract steps. MPEP.04.(d). The claim does not recite an improvement to the functioning of a computer itself or to any other technology or technical field (MPEP § 2106.04(d)(1)). Instead, the claim uses standard AI framework components (a prompt model and LLM) in accordance to their nominal, conventional capabilities (processing natural language to generate output).
The recital of AI/LLM components via specialized terminology such as “prompt model” and “large language model” expressed in a high level of generality about very nominal functions understood rather as a form of a desired result (interpreting inputs, generating requirements and designing solutions) fails to demonstrate/show a particular transformation being accomplished in this computer/SW technology and AI field and as such, these components do not remove the process/method from being an abstract conceptual process. MPEP 2106.04(d)(1)
The “transmitting” and “storing” data are viewed as generic post-solution activities that do not meaningfully limit the claim scope – see MPEP § 2106.04(d)(2))- for they amount to a field-of-use or insignificant extra-solution activity.
Therefore, the above elements fail to integrate the abstract idea into a practical application.
Step 2B
The additional elements recited as “processing device”, “communication device”, “storage device” are generic and well-known components in the relevant computer field, and the “large language model” constitutes standard learning elements being well-understood routine and/or conventional activities in this relevant field. MPEP 2106.05(d).
The “transmitting” and “storing” limitations are viewed as generic post-solution activities that do not meaningfully limit the claim scope – see MPEP § 2106.04(g)- for they amount to a field-of-use or insignificant extra-solution activity
Additional elements recited as prompt model for generating of requirements, training by a LLM to “process requirements for designing SW solution” are viewed as reiterating what the well-understood functionalities nominally amount to and what desired result (requirements, solution) is obtained from using such functionality. No detail is provided to show “how” these functionalities perform to achieve results being “requirements” and “solution”. MPEP 2106.05(a)(b)(c )(d)(f)
The additional elements amount to no more than the Judicial Exception itself.
Construed in an ordered combination, the recited steps follow the workflow of obtaining input[Wingdings font/0xE0]process/analyze it using a model [Wingdings font/0xE0]generate result[Wingdings font/0xE0]transmit/store result. The ordered sequence represents generic computer functionality applied to an Abstract mental process. MPEP 2106.05(e)
In all, claim 1 is ineligible under 35 USC § 101 statute.
Claim 11 is rejected under 35 U.S.C. 101 because the claimed invention is directed to a judicial exception (i.e., a law of nature, a natural phenomenon, or an abstract idea) without significantly more. Claim(s) 11 is/are directed to Abstract Idea. The claim(s) does/do not include additional elements that are sufficient to amount to significantly more than the judicial exception because of the 2-step analysis as follows.
Step I
Claim 11 is directed to a apparatus/system category.
Step 2A
Prong one:
The elements recited as: "obtaining... input data …"; “analyzing.., Markov discloses input data …" and "generating... software solution information... based on the analyzing" are limitations can be performed in the human mind (or by a human using pen and paper) as set forth above in the analysis of claim 1.
Prong two:
The elements recited as “processing device”, “communication device”, “storage device” are disassociated requirements with respect to operation of the Abstract Idea. These are perceived as generic components or hardware construed in a high level of generality and can be viewed as mere conduit to execute the abstract steps. MPEP.04.(d).
Elements recited as “prompt model” and “large language model” expressed in a high level of generality for high-level functions understood rather in form of a desired result (interpreting inputs, generating requirements and designing solutions) fail to show a particular transformation being accomplished in this computer/SW technology and AI field and as such, these components do not remove the process/method from being an abstract conceptual process. MPEP 2106.04(d)(1)
The “transmitting” and “storing” data are viewed as generic post-solution activities that do not meaningfully limit the claim scope – see MPEP § 2106.04(d)(2))- for they amount to a field-of-use or insignificant extra-solution activity.
Therefore, the above elements fail to integrate the abstract idea of claim 11 into a practical application
Step 2B
The additional elements recited as “processing device”, “communication device”, “storage device” are generic and well-known components in the relevant computer field, and the “large language model” constitutes standard learning elements being well-understood routine and/or conventional activities in this relevant field. MPEP 2106.05(d).
The “transmitting” and “storing” limitations are viewed as generic post-solution activities that do not meaningfully limit the claim scope – see MPEP § 2106.04(g)- for they amount to a field-of-use or insignificant extra-solution activity
Additional elements recited as prompt model for generating of requirements, training by a LLM to “process requirements for designing SW solution” are viewed as reiterating what the well-understood functionalities nominally amount to and what desired result (requirements, solution) is obtained from using such functionality. No detail is provided to show “how” these functionalities perform to achieve results being “requirements” and “solution”. MPEP 2106.05(a)(b)(c )(d)(f)
The additional elements amount to no more than the Judicial Exception itself.
Construed in an ordered combination, the recited steps follow the workflow of obtaining input[Wingdings font/0xE0]process/analyze it using a model [Wingdings font/0xE0]generate result[Wingdings font/0xE0]transmit/store result. The ordered sequence represents generic computer functionality applied to an Abstract mental process. MPEP 2106.05(e)
In all, claim 11 is ineligible under 35 USC § 101 statute.
Step 2B analysis of dependent claims.
Claims 2 and 12 recites receiving a pre-trained LLM and training data using domain specific data and computational logic data; and generating LLM based on that training. The input to the pre-trained model and the analyzing of input from prompt data as recited fail to teach limitation that particularly transform the Abstract Idea into a eligible practical application.
Claims 3 and 13 recite embedding information into a pre-trained model based on domain and computational data and finetuning this pre-trained model to generate the LLM based on the fine tuning. As no details are provided excepting describing functionalities in generic and nominal term, the pretraining and finetuning amount to mere well-understood concepts in machine learning field and fail to add significantly more to the abstract Idea.
Claims 4 and 14 recite obtaining, analyzing generating preliminary data, transmitting, receiving comment and generating training data for a LLM based on preliminary data and comment. Mere recital of a function and input thereof cannot establish a non-conventional transformation being accomplished with the Abstract Idea within the field of machine learning.
Claims 5 and 15 recite first and second prompt model, associated with first LLM and second LLM, each coupled to first language or second language model having each a role respectively, where analyzing input data via each prompt model comprises performing a respective programming operation. Mere recital that a function is being performed for a respective model using a form of input does not constitute a improvement to the computer of field of AI associated with the Abstract Idea being set in this AI field.
Claims 6 and 16 recite receiving ecosystem data and training the LLM based thereon. Recital of a training action based on a form of input does not functionally transform the Abstract Idea into a non-standard technical improvement associated with AI computer field.
Claims 7and 17 recite in general terms association between the ecosystem data, a SW platform and software solution. There is no details about how a solution is being achieved and mere listing of input and platform fails to add significantly more in order to transform the Abstract Idea into eligible novelty.
Claims 8 and 18 recite general terms association between a platform, a SW subsystem and a SW solution, without implementation details showing how this association leads to a technical transformation as a way to improve the AI or the Abstract idea in this AI field.
Claims 9 and 19 recite detecting an input, generating data and analyzing a ML model based on the analyzing. All of the activities fall into the Abstract Idea groups characterized by mental process and use of mathematical concepts.
Claims 10 and 20 recite obtaining data associated with a SW solution, using a LLM and prompt model to analyze the data and generate requirements for improving functioning of a SW solution, generating improvement information and transmitting it. The workflow of receiving data, analyzing data and generate information related to an improved solution belongs to a scenario of processing information and generate information for its desired solution, all expressed in high level of generality, none of which indicative of functionalities that particularly add significantly more to the Abstract Idea.
In all, claims 1-20 are deemed non-eligible under the 35 USC § 101 statute.
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.
Claims 1-4, 6-14, 16-20 is/are rejected under § 35 U.S.C. 103 as being unpatentable over Markov et al, USPubN: 2024/0362421 (herein Markov) in view of Rosenstein et al, USPubN: 2020/0005157 (herein Rosenstein) and Lee et al, USPubN: 2023/0359790 (herein Lee)
As per claim 1, Markov discloses a method for facilitating provisioning of software solutions, the method comprising:
obtaining, using a processing device, at least one input data associated with at least one user, wherein the at least one input data comprises at least one natural language input (user-specified natural language – para 0050; content policy … user-specified natural language instructions – para 0031) data associated with at least one natural language;
analyzing, using the processing device, the at least one input data using each of a prompt model (para 0029; Fig. 3; human-curated synthetic data may be generated using prompt template – para 0031) and a large language model (para 0005; large language model – claim 1, pg. 15), wherein the prompt model is coupled with the large language model, wherein the large language model is trained on a training data(cold start data … may comprise human-curated synthetic data … cold start data may be used by a training data engine – para 0031),
wherein the prompt model interprets input data for generating requirements (prompt templates to generate synthetic data – para 0039 – Note1: curating input data (or content policy) into synthetic data which is also subjected to modification, filtering, categorization – see claim 25, pg. 16 -- augmentation, removing and replacement thereof as part of forming model constraints – e.g. taxonomy of content based on input data … input data analyzed, manipulated as a constraint on a machine learning model - para 0051 – and improved format of taxonomy – see para 0058; model constraint – para 0082 - as input into a training, the synthetic data being generated via prompt templates – para 0039- reads on prompt-based synthesis model to interpret/curate user data as cold start data subjected to classification, filtering, modification contributive to generating of requirements/constraints, i.e. the generating thereof via prompt-based formation of data synthetic and categorized taxonomy destined as constraints for a machine learning- para 0005 – or refined language model to implement a solution rooted from this content taxonomy – para 0032 ) of the at least one user, wherein the large language model processes the requirements (see Note1; see para 0051) for designing the software solutions (para 0004-0005; implements a solution – para 0032; taxonomy of content based on input data … input data analyzed, manipulated as a constraint on a machine learning model … implements a solution – para 0051) for the at least one user (see user policy from above);
generating, using the processing device, at least one software solution information for the provisioning of at least one software solution (para 0032, 0051) based on the analyzing (see Note1; para 0082; claim 25, pg. 16) of the at least one input data;
transmitting, using a communication device, the at least one software solution information to at least one device (e.g. transmit output to outcome metrics database, model refinement engine 760 may transmit the received output to featurization engine – para 0083; model output data, validation data … transmit output to ML algorithms database 790 – para 0082 – Note2: model output and validation data thereof sent to a ML database reads on transmitting one solution information to at least one device based on a ML modeling engine and/or a featurization engine – see Engine 720 and 730 - Fig. 7) and
storing, using a storage device, the at least one software solution information (see Note2)
Markov does not explicitly disclose
storing, using a storage device, the at least one software solution information and each of the prompt model and the large language model.
Similar to storing model information, output from a modeling engine and validation data into a ML algorithm database in Markov (Fig. 7) provided also as source for training data, Lee particularly discloses a database (Fig. 1, para 0024) to store component of a design solution per effect of solving design via machine learning (para 0004-0005), the stored data provided as source for training data or UI modeling (para 0030, 0043) or design solution components (para 0035), where solving design problems uses machine learning based on receiving user representative data of a design and categorizing the data into component/function of a machine learning. Hence, software solution information being persisted as DB components representing machine learning generated output for a targeted design solution is recognized
Storing information associated with modeling, machine learning implementation of a design or a SW solution is shown in Rosenstein as a content library database (para 0014, 0022) for containing a plurality of prompts and evaluation data associated with each of the prompts (para 0006), the prompts provided (para 0010-0011, 0013) for identifying an evaluation model, determining a training level/status (para 0018), training the evaluation model, identifying pre-existing response from preexisting model (para 0023-0024) and updating a training based on the model evaluation data (para 0012, 0021) based on respective prompts. Hence, software solution information and each of the prompt model associated with a given training is recognized.
Therefore, based on the bidirectional use of database in Markov (Fig. 7) where previous ML outcome can be either persisted and else, retrieved to other modeling/featurization contexts, it would have been obvious for one of ordinary skill in the art before the effective filing date of the invention to implement the ML database in Markov so that at least one software solution information derived from the modeling environment and featurization engine can be stored therein, in terms of ML output information that also include one or more corresponding prompt model instances – per Rosenstein library database - and data obtained from executing large language model – per the design solutions databased in Lee; because
insights or learning details obtained from persisted solution data or output from previous LLM execution/inferencing can be used toward future problem resolving projects/developments and persisted DB record provided as recommendation insights, contextual question/answer, applicable conditions, requirement/constraint from previously generated or customized prompt instances collected from past prompt-based model analyses associated with a LLM such as in Markov framework, can serve as configuration aid or support into a new customization of machine learning that relies on prompt-template methodology as set forth in Markov, in that the DB record reuse would mitigate effort of a developer and reduce payload, modeling resource in constructing a new machine learning instance of modeling/pipeline geared for seeking a best software solution or proposed implementation for problem solving software.
As per claim 2, Markov discloses method of claim 1 further comprising:
retrieving, using the storage device (database 790 … store one or more machine learning models … may include a GNN, a GPT (pre-trained transformer - para 0082), a pre-trained large language model (accessing a pre-trained language model – claim 1 pg. 15; claim 11, pg. 16);
training, using the processing device, the pre-trained large language model (see generating training – claim 1 and claim 16 pg. 15-16) using the training data, wherein the training data comprises each of a domain data (e.g. generating training data based on the multi-domain cold start data – claim 1, pg. 15) specific to a domain (para 0084-0085; public domain data – para 0033; WDAT – para 0037; content policy – para 0060; content policy 101b, content policy may include user-specified natural language – para 0031-0032; Fig. 1) and a computational logic data (pre-trained language model based on the input data and the training data - claim 1, pg. 15; model may be be pre-trained – para 0034 - Note3: model being pre-trained on corpus text and input data reads on computation logic encoded with a pre-training logical instance) associated with a computational logic; and
generating, using the processing device, the large language model (generate an optimized language model - claim 11, pg. 6; generate an optimized language model – claim 1, pg. 15) based on the training (see above), wherein the analyzing of the at least one input data using each of the prompt model (para 0029; Fig. 3; synthetic data may be generated using prompt template – para 0031) and the large language model (refer to claim 1) is based on the generating of the large language model.
As per claim 3, Markov discloses method of claim 2, wherein the training of the pre-trained large language model comprises embedding at least one information (access a language model … from a remote … model storage … based on desired output behaviors … which may be encoded … model may be pre-trained … the accessed language model … decoder model … language model based on … content policies … user-defined model parameters, content classification, annotated data, labeled data – para 0034 - ) into a model architecture of the pre-trained large language model based on at least one of the domain data (e.g. para 0084-0085; public domain data – para 0033; WDAT – para 0037; content policy – para 0031-0032) and the computational logic data (see Note3),
wherein the training of the pre-trained large language model further comprising fine tuning the pre-trained language (fine-tuning a language model from …access engine 108- para 0041; accessing a pre-trained language model – claim 1 pg. 15; claim 11, pg. 16) using at least one of the domain data (refer to claim 3) and the computational logic data based on the embedding (refer to claim 3), wherein the generating of the large language model is further based on the fine tuning of the pre-trained language model (fine tuning a language model from ..model access engine – para 0060; accessing a pre-trained language model – claim 1 pg. 15; claim 11, pg. 16).
As per claim 4, Markov discloses method of claim 2, wherein the large language model is associated with a domain (refer to claim 2), wherein the method further comprises:
obtaining, using the processing device, at least one first data associated (content policy may include … instruction, defined task or any combination of parameters that set one or more constraints of a language model – para 0031) with the domain (refer to claim 2);
analyzing, using the processing device, the at least one first data using at least one first large language model (cold start data may be used to initiate an active learning and labelling process for a language model - para 0031);
generating, using the processing device, at least one preliminary data (unlabeled data or raw data – para 0033; initial configuration … can change over time – para 0027) using the at least one first large language model based on the analyzing of the at least one first data (see above);
transmitting, using the communication device, the at least one preliminary data to at least one domain expert device (training data generation engine 106 may receive … raw data – para 0033) of at least one domain expert (refer to domain data and content policy trained by language model per claims 2-3);
receiving, using the communication device, at least one comment of the at least one domain expert from the at least one domain expert device (Fig. 2; taxonomy generated may comprise content categories … derogatory stereotypes … neutral statement … quote of other individual to provide commentary … sub-categorical layer may be ranked by a metric … metric may be automatically generated based on training data or cold start data … sub-categories within the taxonomy generation implements a solution – para 0051; Fig. 5 – Note4: taxonomy provided into a training environment that includes commentary category as one of the sub-categories domains being categorized or ranked as metrics forming input into data generation engine – engine 106, Fig. 1 - for a training model – Fig. 5 - reads on expert domain assessing system sending a comment to be handled by a domain expert training system that includes a data generation engine and a training model); and
generating, using the processing device, the training data for the large language model based on the at least one preliminary data (see above) and the at least one comment (see Note4 from above), wherein the training of the large language model (refer to claim 1) using the training data is based on the generating (engine 106 – Fig. 1) of the training data.
As per claims 6-7, Markov discloses method of claim 1 further comprising:
(i) receiving, using the communication device, a software ecosystem data of a software ecosystem of the at least one user (content policy – para 0031; content taxonomy – 507 – Fig. 5; receive one or more content policies … taxonomy generation engine – para 0032; multi-domain data, human-curated synthetic data – para 0031; user-specified natural language – para 0050) from at least one user device; and training, using the processing device (e.g. para 0041-0042) the large language model (language model, step 511, 513 – Fig. 5; Predictive output generation engine – Fig. 7) based on the software ecosystem data (see multi-domain, content policy from above – Note5: a API-equipped system/environment – para 0029 – provided with input conversion/synthesis capability – para 0031 - to parse and interpret user contextual data, policies or business intents, sub-categories – para 0032, 0034, 0045 - via use of transform/extraction engine, I/O communication and algorithms – ML algorithms- Fig. 7 - by which training of taxonomy, cold start, synthetic data enable training outputs to be validated – Fig. 5 - and fed back into the cyclical stages – para 0061 - that perform fine-tuning and yielding a solution – para 0041-42 - reads on a SW ecosystem whose received input comes from user policy, labelled data, and multi-domain contextual language – see para 0034, 0036, 0041 – which is equipped with engine, APIs for transforming the input data and training thereof to implement, improve a SW solution – para 0027, 0032, 0051 - for a given user domain of interest and),
wherein the generating of the at least one software solution information is further based on the training of the large language model (see above; iteratively training … of the language model - para 0042 ) based on the software ecosystem data (refer to Note5).
(ii) wherein the software ecosystem data is associated with at least one of at least one software system and at least one software platform (para 0023, 0081; Fig. 5, 7; see Fig. 2 ) associated with the at least one user (see Fig. 5, Fig. 2; user content policy from above; categories, S1, S2, S3 content, targeting specific chosen groups because of their identities, H, H1, H2 category, V, V1, V2 categories, metric may be automatically generated based on the training, ranking of generated content categories or sub-categories … implements a solution rooted in computer technology … contribute to solving the complex problem … across a multiple type of categories – para 0032),
wherein the at least one software solution comprises at least one software subsystem for at least one of the at least one software system (based on the training, ranking of generated content categories or sub-categories … implements a solution rooted in computer technology … contribute to solving the complex problem … across a multiple type of categories – para 0032) and the at least one software platform (see Fig. 1, 7) , wherein the at least one software solution (improv - para 0032, 0051 ) is for the at least one software subsystem.
As per claim 8, Markov discloses method of claim 1, wherein the at least one software solution comprises at least one of at least one software system (refer to claim 7), at least one software platform (e.g. para 0023, 0081; Fig. 5, 7; see Fig. 2), and at least one software subsystem for at least one of the at least one software system and the at least one software platform, wherein the at least one software solution information is associated with at least one of the at least one software system, the at least one software platform, and the at least one software subsystem (refer to claims 6-7).
As per claim 9, Markov discloses method of claim 1 further comprising:
detecting, using at least one input device, at least one input of the at least one user (user-specified natural language – para 0050), wherein the at least one input comprises at least one of an utterance (AI systems to interpret …. Amounts of information … recognize objects, natural language, human speech – para 0028) and an image (image data or video data – para 0031; input data may comprise … text data, image data – para 0050);
generating, using the processing device, at least one data (instruction to filter out content data; instruction may correspond to a particular language model application network … digital pattern, format, style – para 0050) based on the detecting;
analyzing, using the processing device, the at least one data using at least one first machine learning model (particular language model – para 0050; LM 108 - Fig. 1), wherein the at least one first machine learning model converts the data to text data (cold start data 101, human curated synthetic data – para 0050; template – Fig. 3); and
generating, using the processing device, the at least one input data based on the analyzing of the at least one data (see above), wherein the obtaining of the at least one input data comprises the generating of the at least one input data (synthetic data – para 0050; Fig. 3) based on the analyzing (para 0050) of the at least one data.
As per claim 10, Markov discloses method of claim 1, wherein the at least one software solution is deployed based on the at least one software solution information, wherein the method further comprising:
obtaining, using the processing device, at least one functioning data (taxonomy, synthetic data, cold start data per claim 1) associated with a functioning of the at least one software solution (taxonomy implements a solution - para 0032);
analyzing, using the processing device (data transformation engine, featurization engine, modeling engine – Fig. 7), the at least one functioning data using each of the prompt model (refer to claim 1) and the large language model (refer to claim 1),
wherein the prompt model interprets functioning data for generating requirements (refer to claim 1; see Note1) for improving the functioning of the at least one software solution (para 00, wherein the large language model processes (refer to claim 1; para 0005; Fig. 5) the requirements (see Note1) for generating improvements (para 0027, 0032, 0051 ) for the at least one software solution;
generating, using the processing device, at least one improvement information (para 0027, 0032, 0051) for improving the functioning of the at least one software solution(refer to claim 6-7) based on the analyzing of the at least one functioning data (Fig. 5); and
transmitting, using the communication device, the at least one improvement information to the at least one device (refer to claim 1; refer to Note2).
As per claim 11, Markov discloses a system for facilitating provisioning of software solutions, the system comprising:
a processing device configured for:
obtaining at least one input data associated with at least one user, wherein the at least one input data comprises at least one natural language input data associated with at least one natural language;
analyzing the at least one input data using each of a prompt model and a large language model, wherein the prompt model is coupled with the large language model, wherein the large language model is trained on a training data,
wherein the prompt model interprets input data for generating requirements of
the at least one user, wherein the large language model processes the requirements for designing the software solutions for the at least one user; and
generating at least one software solution information for the provisioning of at least one software solution based on the analyzing of the at least one input data;
( All of which having been addressed in claim 1)
a communication device (refer to transmit in claim 1; para 0076-0077) communicatively coupled with the processing device, wherein the communication device is configured for transmitting (refer to claim 1) the at least one software solution information to at least one device; and a storage device (refer to claim 1; para 0076-0077) communicatively coupled with the processing device,
wherein the storage device (see above) is configured for storing model (refer to rationale A of claim 1) the at least one software solution information and each of the prompt model and the large language
As per claim 12, refer to rejection of claim 2.
As per claim 13, refer to rejection of claim 3.
As per claim 14, refer to rejection of claim 4.
As per claims 16-17, refer to rejection of claims 6-7.
As per claim 18, refer to rejection of claim 8.
As per claim 19, refer to rejection of claim 9.
As per claim 20, refer to rejection of claim 10.
Claims 5, 15 is/are rejected under § 35 U.S.C. 103 as being unpatentable over Markov et al, USPubN: 2024/0362421 (herein Markov) in view of Rosenstein et al, USPubN: 2020/0005157 (herein Rosenstein) and Lee et al, USPubN: 2023/0359790 (herein Lee) further in view of McCarthy, USPN: 11,765,207(herein McCarthy)
As per claim 5, Markov discloses method of claim 1, wherein the prompt model comprises a first prompt model and a second prompt model, wherein the large language model comprises a first language model (prompt templates – para 0046) and a second language model (prompt templates – para 0046), wherein the first prompt model is coupled with the first language model (prompt template may be crafted by a human operator, user-crafted templates – para 0046), wherein the second prompt model is coupled with the second language model (Note5: prompt templates being each crafted by a human reads on first and second prompt model construed as first or second language model, each language expressing a human operator content or context/category that may be one indicative of hateful, violence, non-erotic, or particularly contextualized or undesired sub-category – Fig 2; para 0045)
wherein the analyzing of the at least one input data (para 0045; Fig. 2) using the each of the prompt model (see above) and the large language model (Fig. 5) comprises
performing at least one first programming operation by each of the first prompt model (see prompt templates from above) and the first large language model (see Note5 from above) and performing at least one second programming operation by each of the second prompt model (see prompt templates from above) and the second large language model (see Note5 from above)
Markov does not explicitly disclose analyzing of the at least one input data using the each of the prompt model and the large language model in terms of
wherein the method comprises assigning, using the processing device, a first role for each of the first prompt model and the first large language model, and a second role for each of the second prompt model and the second large language model,
performing at least one first programming operation by each of the first prompt model and the first large language model based on the first role of each of the first prompt model and the first large language model, and performing at least one second programming operation by each of the second prompt model and the second large language model based on the second role of each of the second prompt model and the second large language model.
McCarthy discloses configuration report based on large language model configured with prompt dataset based on policy statement (Fig. 13) set into prompt template (col. 4 li. 64 to col .5 li. 17), where the LL model is trained on large collection of NL documents and generates predictive response based on text prompts (col. 4, li. 17-37) including generating and validating candidate configuration profile (col.5 li. 18-31) that includes names and field values associated the management of the network resources by which to validate the profile (Fig. 14), and providing the validated configuration to a infrastructure security computer (ISC) so that network policies can be updated based thereon (col. 5 li. 3-17, li. 48-58), where declaring network policies using prompt template include characteristics of a resource and user role (col. 6 li. 1-9), where the configuration profile employed based on such characteristics (users, user role … configuration profiles may be employed for users, network, sub-networks declare or define … network policy for users, network - col. 17 li. 22-36) and declared policy statement is provided via a match process (col. 5 li. 39-47; col. 6 li. 10-23); e.g. reporting the mismatch based on which to ensure security/protection of access into the network. Hence role assigned to user included into configuration of prompt template and declaration of network policies (see Abstract) for a LLM to predict responses by which to implement security protection to resources of a NW is recognized.
Therefore, based on policy data being a particular user context or domain provided via a prompt template in Markov, it would have been obvious for one of ordinary skill in the art before the effective filing date of the invention to implement formation of prompt model for each user domain-specific context or policy domain so that analyzing of the at least one input data using the each of the prompt model and the large language model would include
assigning, using the processing device, a first role for each of the first prompt model and the first large language model, and a second role for each of the second prompt model and the second large language model – as shown above in McCarthy incorporating user role into policy and prompt data;
performing at least one first programming operation by each of the first prompt model and the first large language model based on the first role of each of the first prompt model and the first large language model – as set forth per McCarthy declaring of policy and formation of prompt templates;
and/or performing at least one second programming operation by each of the second prompt model and the second large language model based on the second role of each of the second prompt model and the second large language model – as set forth per the user role configuration and prompt data generated therefor in McCarthy configuration of LLM; because
use of a LLM training in conjunction with a integration platform that operates from user policy input spanning a multi-domain content/context for which a solution is to be attained via coordination by the platform with cycles of a machine learning training on basis of data obtained from extraction and synthesis of the initial user/domain input into corresponding prompt as set forth in Markov approach, in that prompt data integrated with the user assigned role can impart a criteria or weight into the training cycles would enable a SW solution when attained to significantly respond to or suit well with to the user intent, concern, context or domain of interest in direct relevance with a given role or user registered capacity within the enterprise; therefore boosting usability of prompt-based training framework in that as configured, the SW solution attained based on LLM and prompt data by the framework would more accurately address/resolve a domain-specific field of concern or domain policy which is affecting a target user in a given role.
As per claim 15, refer to rationale of claim 5.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Tuan A Vu whose telephone number is (571) 272-3735. The examiner can normally be reached on 8AM-4:30PM/Mon-Fri.
If attempts to reach the examiner by telephone are unsuccessful, the examiner's supervisor, Chat Do can be reached on (571)272-3721.
The fax phone number for the organization where this application or proceeding is assigned is (571) 273-3735 ( for non-official correspondence - please consult Examiner before using) or 571-273-8300 ( for official correspondence) or redirected to customer service at 571-272-3609.
Any inquiry of a general nature or relating to the status of this application should be directed to the TC 2100 Group receptionist: 571-272-2100.
/Tuan A Vu/
Primary Examiner, Art Unit 2193
Septembre 01, 2026
Prosecution Insights