DETAILED ACTION
Introduction
1. A response was filed in this application on 04/15/2026 after the non-final rejection of 01/16/2026 and the Applicant initiated interview of 04/14/2026. Claims 1-20 are amended while no claims are cancelled or added in this latest submission by the Applicant. Thus, claims 1-20 are currently pending for reconsideration by the Examiner and are examined below. 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
2. The rejection under 35 U.S.C. 101 is withdrawn for claims 9-14 in light of the Applicant’s amendments/arguments submitted in this latest response coupled with the disclosure in Applicant’s specification of paragraphs 8-9, 42-45 and further in view of the limitations pertaining to storing the plurality of at least one query variation-answer pair in a semantic cache along with providing an answer to a subsequent query from the semantic cache using the at least one query variation-answer pair.
For claims 1-8 and 15-20, the rejection under 35 U.S.C. 101 is maintained. The Applicant’s arguments in this regard have been fully considered but are unpersuasive. With regards to the Applicant’s arguments for step 2A, prong 1 of the analysis under 35 U.S.C. 101, the Applicant quotes certain sections of paragraphs 8-9, however the independent claims 8 and 15 do not recite these sections. Further, the independent claims do not recite selection, crossover and mutation operation in the manner which would make them substantially more than a mental process. Similarly, with regards to the Applicant’s arguments for step 2A, prong 2 of the analysis under 35 U.S.C. 101, the features that are argued by the Applicant as recited in paragraphs 8-9, 42 (or even 45) are not recited in the independent claims 8 and 15 in a manner which would indicate an improvement in the technological field. The Applicant is advised to bring in limitations from the independent claim 9, especially the ones pertaining to storing the plurality of at least one query variation-answer pair in a semantic cache and providing an answer to a subsequent query from the semantic cache using the at least one query variation-answer pair.
With regards to the prior art of record, the Applicant’s arguments have been fully considered but are unpersuasive for at least the reasons outlined below.
The Applicant argues that Examiner quoted prior art reference Larson (U.S. Patent Application Publication # 2025/0328565 A1) does not describe a genetic algorithm in any capacity. Applicant further argues that Larson provides "iterative graph-walking and embedding-change comparison" which does not correspond to a genetic algorithm for producing variations of a query.
The Examiner respectfully disagrees and argues that the Applicant has not provided any argument beyond a general allegation on why and how Larson’s "iterative graph-walking and embedding-change comparison" does not correspond to a genetic algorithm for producing variations of a query. In absence of such a rationale, the Applicant’s arguments fail to comply with 37 CFR 1.111(b) because they amount to a general allegation that the claims define a patentable invention without specifically pointing out how the language of the claims patentably distinguishes them from the references. Further, Larson teaches a genetic algorithm in paragraphs 27, 232 and 250 wherein machine learning algorithms or models can be implemented in RAG systems by way of crypto-processors which are specialized processors that execute cryptographic algorithms within hardware.
Applicant also contends that Larson does not provide for any form of selection, crossover, and mutation of candidate solutions across generations.
The Examiner respectfully disagrees once again and argues that the instant claims nowhere recite “crossover” and “candidate solutions across generations”. In response to applicant's argument that the references fail to show certain features of the invention, it is noted that the above-mentioned features upon which applicant relies are not recited in the rejected claims. Although the claims are interpreted in light of the specification, limitations from the specification are not read into the claims. See In re Van Geuns, 988 F.2d 1181, 26 USPQ2d 1057 (Fed. Cir. 1993). The Applicant is welcome to make these features a part of the independent claims with suitable amendments if they are supported by their as-filed specification. However, insofar as crossover and mutation is concerned, Larson in para 218, teaches use of cross-entropy error between the output and the ground-truth annotation, while para 168, teaches mutation with regards to a graph process in terms of leaves, nodes, branches and para 177, discloses a cross-encoder.
Applicant thereafter argues that Larson does not describe generating "query variations" at all, and rather, processes individual plain text questions and retrieves relevant text blocks from a document-graph repository. Larson thus does not generate variations of queries.
The Examiner respectfully disagrees once again and argues that Applicant has not address the specific findings of the Examiner from the explicit teachings of the cited prior art. For example, with regards to query variations, the Examiner had referred to para 36 of Larson, which teaches RAG-AI boosting via composition of adjacent context-tagged text blocks via iterative graph-walking and embedding-change comparison. Further variation of queries can be obtained by the graph methodology of linking and concatenation as outlined in para 37 of Larson.
Applicant thereafter argues that the relevance scores described in Larson are fundamentally different from the claimed "fitness score". According to the Applicant, Larson's relevance scores are assigned to retrieved text blocks (i.e., document content) to measure how relevant each text block is to a given question. In contrast, as per the Applicant, the claimed fitness score evaluates query variations (i.e., it scores how well different query formulations perform in retrieving relevant content). Applicant alleges that Larson scores documents for relevance to a query, whereas the claims score query variations for fitness within a genetic algorithm optimization loop.
The Examiner respectfully disagrees once again and argues that the claims nowhere recite “an optimization loop”. The Applicant is welcome to make these features a part of the independent claims with suitable amendments if they are supported by their as-filed specification. Further, Applicant once again has not provided any argument (beyond a general allegation) on why and how Larson’s relevance score is not the same as the claimed fitness score. Applicant also fails to outline how the claimed query variations are in any way different from the previously claimed query-answer pairs. A broad but reasonable interpretation of the claimed query variations can be that they are nothing more that the previously claimed question answer pairs. In light of such an interpretation, the metes and bounds of the claimed limitations with regards to fitness score are met by the relevance score of Larson.
Finally, the Applicant alleges that Larson does not provides for "processing subsequent queries to the LLM based on the at least one query variation- answer pair." Larson, as per the Applicant, processes each new question through the full RAG pipeline (search, re-rank, generate) rather than using pre-generated query variation-answer pairs to answer subsequent queries.
The Examiner respectfully disagrees once again and argues that the claims nowhere recite “using pre-generated query variation-answer pairs to answer subsequent queries”. The Applicant is welcome to make these features a part of the independent claims with suitable amendments if they are supported by their as-filed specification. Further, Larson is also teaching an optimized loop by way of an iterative graph-walking and embedding-change comparison along with optimized re-ranking (Paragraphs 36 and 219).
The Applicant has not presented any other arguments for the remaining claims and those are therefore also deemed addressed by the discussion above. The prior art rejection is therefore sustained.
Information Disclosure Statement
3. The Information Statement (IDS) filed on 04/15/2026 has been accepted/considered and is in compliance with the provisions of 37 CFR 1.97.
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.
4. Claims 1-8 and 15-20 are rejected under 35 U.S.C. 101 as being nothing more than an abstract idea. As an example, regarding claim 1, the limitations of obtaining question-answer pairs, processing them by modifying them, identifying content portions within them, scoring them and generating new question-answer pairs from the aforementioned processing fall under the category of mental processes. These steps are drafted at a high level of generality without tying it to a specific technological improvement. More specifically, these steps can be performed in the mind of a human being with at most the aid of a pen and paper but for the recitation of generic computer components, and thus it falls within the -Mental Processes- grouping of abstract ideas. Accordingly, this claim recites an abstract idea.
This judicial exception is not integrated into a practical application because the
recitation of a device, a system (including a RAG system), processor, AI model and/or a computer readable medium merely read to generalized computer components, based upon the claim interpretation wherein the structure is interpreted using the specification.
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to the integration of the abstract idea into a practical application, the additional element of using generalized computer components to generate, extract, determine, and generate, amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claim is therefore not patent eligible.
Claims 2-8, only provide certain details of the mental processes outlined above, such receiving a subsequent query, training the RAG system, identifying subsets of query variations, mixing/mutating tokens and processing user feedback etc. These are all steps which themselves can also be accomplished by a human being with at most the aid of a pen and paper and hence also do not amount to significantly more than the judicial exception. Claims 15-20, are computer readable medium (CRM) claims corresponding to method claims 8-14 and hence are also rejected at least for the reasons outlined above.
Claim Rejections - 35 USC § 102
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(2) The claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.
5. Claims 1-7, 9-13 and 15-20 are rejected under 35 U.S.C. 102 (a) (1) as being anticipated by Larson (U.S. Patent Application Publication # 2025/0328565 A1).
With regards to claim 1, Larson teaches a computer-implemented method, executed on a computing device, comprising identifying a query-answer pair associated with a generative artificial intelligence (AI) model comprising a large language model (LLM) (Para 27, teaches a user of a scientific instrument who can have inquiries regarding how the scientific instrument should be operated, maintained, serviced, or troubleshot. In various aspects, such inquiries can be automatically answered by leveraging generative artificial intelligence. In particular, such inquiries can be automatically answered by leveraging retrieval augmented generative artificial intelligence or RAG-AI. This RAG-AI can involve a large language model or LLM);
generating, using a genetic algorithm, a first plurality of query variations of a query of the query-answer pair (Para 36, teaches RAG-AI boosting via composition of adjacent context-tagged text blocks via iterative graph-walking and embedding-change comparison. Para 37, further teaches generating a graph whose nodes and edges collectively represent the contents and hierarchical structure of the given technical document. In particular, leaf nodes of such graph can represent individual text blocks written in the given technical document, and non-leaf nodes of such graph can represent titles, section or subsection headings, page numbers, instrument identifiers, dates, or any other contextual information that are written in the given technical document and that encompass, encapsulate, apply to, or otherwise qualify respective text blocks of the given technical document. In various cases, each leaf node of such graph can be considered as being linked or concatenated with whatever non-leaf nodes from which it depends. In other words, any text block within such graph can be considered as being tagged with whichever titles, section or subsection headings, page numbers, instrument identifiers, dates, or other contextual information below which that text block is nested in such graph);
identifying, from a set of reference documents, a plurality of content portions associated with the first set of query variations using a Retrieval Augmentation Generation (RAG) system associated with the LLM (Para 36, teaches RAG-AI boosting via composition of adjacent context-tagged text blocks via iterative graph-walking and embedding-change comparison. Para 38, further teaches that if an embedding of that composed context-tagged text block differs by less than any suitable threshold margin from an embedding of the relevant context-tagged text block and if that composed context-tagged text block does not violate token limits of the LLM, then the composed context-tagged text block can be used to answer the given natural language question. In contrast, if the embedding of that composed context-tagged text block differs by more than the threshold margin from the embedding of the relevant context-tagged text block or if that composed context-tagged text block violates token limits of the LLM, then the relevant context-tagged text block can be used to answer the given natural language question, rather than the composed context-tagged text block);
determining a fitness score associated with each of the query variations of the first plurality of query variations using based on content portions the plurality of content portions identified based on corresponding query variations (Para 36, further teaches text block re-ranking based on synthesized responses to chain-of-thought prompts. Para 39, further teaches that in order to reconcile or compare these differently-discovered context-tagged text blocks, a re-ranker can be implemented to assign to each discovered or found context-tagged text block a relevance score showing how relevant or irrelevant a respective context-tagged text block is to the given natural language question. Para 177, teaches that the re-ranker can be a cross-encoder based on bi-directional encoder representations from transformers);
selecting, using the genetic algorithm, at least one query variation from the query variations of the first plurality of query variations based on the fitness score associated with each query variation of the first plurality of query variations (Para 81, teaches selecting, by the device and via execution of a re-ranker, one or more highest-ranking context-tagged text blocks from the plurality of context-tagged text blocks; and generating, by the device, a unified prompt by concatenating the plain text question with the one or more highest-ranking context-tagged text blocks, wherein the large language model can receive the unified prompt as input and can produce the plain text answer as output. The device can cause the large language model to respond to one or more chain-of-thought prompts for respective ones of the plurality of context-tagged text blocks, thereby yielding a plurality of chain-of-thought responses that respectively correspond to the plurality of context-tagged text blocks, the re-ranker can assign respective relevance scores to the plurality of chain-of-thought responses, and the device can identify the one or more highest-ranking context-tagged text blocks based on the relevance scores);
and generating at least one query variation-answer pair based on the at least one query variation (Para 36, further teaches a repository or database of document-graphs, each document-graph comprising respective context-tagged text blocks and prompt augmentation for identifier emphasis. Para 98, teaches generating a unified prompt, by concatenating those multiple top or most-relevant context-tagged text blocks with the plain text question);
and processing subsequent queries to the LLM based on the at least one query variation-answer pair (Para 106, provides examples of multiple subsequent queries).
With regards to claim 2, Larson teaches the computer-implemented method of claim 1, further comprising providing an answer to at least one of the subsequent queries from a semantic cache using the query variation-answer pair (Para 98, teaches a document-graph can represent the semantic or substantive structure and content of a respective technical document. For any potentially-relevant context-tagged text block, the second logic can involve generating a composed context-tagged text block, via iterative graph-walking and embedding-change comparison with respect to that potentially-relevant context-tagged text block).
With regards to claim 3, Larson teaches the computer-implemented method of claim 1, further comprising training the RAG system using the at least one query variation-answer pair (Para 214, teaches that when it is desired to train the re-ranker, the training input can be any suitable pair of training texts e.g., a training question and a training text block; a training question and a set of training chain-of-thought responses, and the ground-truth annotation can be whatever correct or accurate relevance score that is known or deemed to correspond to the training input).
With regards to claim 4, Larson teaches the computer-implemented method of claim 1, wherein generating the first plurality of query variations includes tokenizing the query of the query-answer pair into a plurality of tokens (Para 28, teaches breaking up lengthy technical documents into smaller, discrete blocks of text e.g., into individual pages, sections, paragraphs, or passages so as to comply with token limits of the LLM).
With regards to claim 5, Larson teaches the computer-implemented method of claim 4, wherein generating the at least one query variation-answer pair includes selecting a subset of the first plurality of query variations using the fitness score associated with each of the query variations of the first plurality of query variations (Para 72, teaches that the search component can compute a respective relevance score for each potentially-relevant context-tagged text block. The search component can identify a subset of the plurality of potentially-relevant context-tagged text blocks that are actually relevant to the plain text question, based on those relevance scores).
With regards to claim 6, Larson teaches the computer-implemented method of claim 5, wherein generating the at least one query variation-answer pair includes mixing tokens from the subset of the first plurality of query variations to generate a second plurality of query variations (Para 168, teaches that the search component can concatenate, aggregate, combine, or otherwise compose that nearest or most adjacent text block with the specific text block, with all of the contextual information e.g., non-leaf nodes, that is upstream of the specific text block, and with all of the contextual information that is upstream of that nearest or most adjacent text block. Such concatenation can be referred to as the composed context-tagged text block. Furthermore, the search component can generate an embedding for the composed context-tagged text block, by passing the composed context-tagged text block through the encoder portion).
With regards to claim 7, Larson teaches the computer-implemented method of claim 6, wherein generating the at least one query variation-answer pair includes mutating randomly selected tokens from the subset of the first plurality of query variations to generate the second plurality of query variations (Para 168, further teaches that the search component can generate a composed context-tagged text block, by iteratively performing graph-walking and embedding-change comparison with respect to the potentially-relevant context-tagged text block. More specifically, the potentially-relevant context-tagged text block can include one leaf node that represents a specific text block within the document-graph. The search component can traverse or walk along the edges of the document-graph so as to identify whichever other text block e.g., other leaf node, within the document-graph is nearest or most adjacent to that specific text block e.g., is separated from that specific text block by a minimum number of intervening levels, branches, or nodes. In case two or more other text blocks are equidistant from the specific text block, the search component can select randomly from among those two or more other text blocks. Para 218, teaches that an error e.g., mean absolute error, mean squared error or cross-entropy error between the output and the ground-truth annotation can be computed. The trainable internal parameters of the artificial intelligence model can be incrementally updated via backpropagation e.g., stochastic gradient descent based on the error).
With regards to claims 9-13, these are system claims for the corresponding method claims 1-7. These two sets of claims are related as method and apparatus of using the same, with each claimed system element's function corresponding to the claimed method step. Accordingly, claims 9-13 are similarly rejected under the same rationale as applied above with respect to method claims 1-7. Larson also teaches storing the plurality of at least one query variation-answer pair in a semantic cache (Para 50) along with providing an answer to a subsequent query from the semantic cache using the at least one query variation-answer pair (Para 45).
With regards to claims 15-20, these are CRM claims for the corresponding method claims 1-7. These two sets of claims are related as method and CRM of using the same, with each claimed CRM element's function corresponding to the claimed method step. Accordingly, claims 15-20 are similarly rejected under the same rationale as applied above with respect to method claims 1-7.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
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.
6. Claims 8 and 14 are rejected under 35 U.S.C. 103 as being unpatentable over Larson in view of Durg (U.S. Patent Application Publication # 2025/0173330 A1).
With regards to claim 8, Larson may not explicitly detail the limitation comprising processing user feedback associated with the at least one query variation-answer pair. However, Durg teaches this (See para 33).
Larson and Durg can be considered as analogous art as they belong to a similar field of endeavor in retrieval augmented generation systems (See Durg, para 40). It would thus have been obvious to one having ordinary skill in the art to advantageously combine the teachings of Durg (Use of user feedback for question-answer pairs) with those of Larson (Use of RAG-Ai system for question-answer boosting) so as to fine-tune one or more of the plurality of LLM models (Durg, para 33).
With regards to claim 14, this is a system claim for the corresponding method claim 8. These two claims are related as method and apparatus of using the same, with each claimed system element's function corresponding to the claimed method step. Accordingly, claim 14 are similarly rejected under the same rationale as applied above with respect to method claim 8.
Conclusion
7. THIS ACTION IS MADE FINAL. The Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a). The following prior art, made of record but not relied upon, is considered pertinent to applicant's disclosure: Puttagunta (U.S. Patent Application Publication # 2025/0321959 A1), Klein (U.S. Patent Application Publication # 2025/0307238 A1). These references are also included in the PTO-892 form attached with this office action.
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any extension fee pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the date of this final action.
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.
Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pair-direct.uspto.gov. If you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). In case you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to NEERAJ SHARMA whose contact information is given below. The examiner can normally be reached on Monday to Friday 8 am to 5 pm. If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Pierre Louis-Desir can be reached on 571-272-7799 (Direct Phone). The fax number for the organization where this application or proceeding is assigned is 571-273-8300.
/NEERAJ SHARMA/
Primary Examiner, Art Unit 2659
571-270-5487 (Direct Phone)
571-270-6487 (Direct Fax)
neeraj.sharma@uspto.gov (Direct Email)
Prosecution Insights