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 Amendment
No claims have been amended. Claims 1-23 are pending in the application.
Response to Arguments
Claim Rejections - 35 USC § 102
In response, the 35 U.S.C. 102 rejection of claims 1-23 has been withdrawn. However, Examiner relies on a new combination of references.
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, 11-16 and 23 are rejected under 35 U.S.C. 103 as being unpatentable over Bosnjakovic (US 2025/0390516 Al) in view of MASSIE (US 2025/0272581 Al)
Regarding claim 1, Bosnjakovic discloses: A method for improving generative artificial intelligence (Al) software application response, comprising: receiving a query directed to a generative Al software application; (Bosnjakovic [0004] receiving… a query including a plurality of sub-queries; [0040] the LLMs… receive queries…the LLMs 125 can form part of one or more generative AI models)
receiving a response to the query, the response generated by the generative Al software application; (Bosnjakovic [0040] The LLMs 125 may include one or more LLMs that are configured to generate responses to user queries or sub-queries)
generating a first contextual value based on the received query;
(Bosnjakovic, [0042] determine a context for each of the identified Queries; [0048] generate a plurality of second vectors indicative of the corresponding sub-queries [0058] second vector indicative of the context of the corresponding sub-query;
generating a second contextual value based on the received response;
(Bosnjakovic, [0048] generate a first vector indicative of the response; [0058] generate a first vector indicative of the response summary ; [0057] the summarized responses are consistent (in context) with their corresponding sub-queries. [0022] comparing each response with its corresponding sub-query and/or its context and may ensure that the context of each response is consistent with the context of its corresponding sub-query;)
,
generating a verification score, (Bosnjakovic [0058] compare a respective response summary with the context for the corresponding sub-query to generate a similarity score,… can generate a first vector indicative of the response summary and generate a second vector indicative of the context of the corresponding sub-query, and then generate the similarity score (corresponding to “a verification score ) based on a cosine distance between the first vector and the second vector.)
based on a value related to a semantic similarity between the received query and the received response, (Bosnjakovic, [0048] the similarity engine 127 can generate a first vector indicative of the response and generate a plurality of second vectors indicative of the corresponding sub-queries. The similarity engine 127 can determine
the similarity score for a respective agent 130 based on a cosine distance between the first vector and the second vector associated with the respective agent [0058] compare a respective response summary with the context for the corresponding sub-query to generate a similarity score,… can generate a first vector indicative of the response summary and generate a second vector indicative of the context of the corresponding sub-query, and then generate the similarity score based on a cosine distance between the first vector and the second vector.)
based on the first contextual value and the second contextual value; (Bosnjakovic [0058] compare a respective response summary with the context for the corresponding sub-query to generate a similarity score,… can generate a first vector indicative of the response summary and generate a second vector indicative of the context of the corresponding sub-query, and then generate the similarity score based on a cosine distance between the first vector and the second vector.)
and initiating a mitigation action in response to detecting that the verification score is below a predetermined threshold. (Bosnjakovic [0058] then compare the similarity score with a threshold to determine whether the respective response summary is sufficiently similar (in context) with the corresponding subquery to be included as part of the natural language answer…if the similarity score for a respective response summary is not greater than the threshold, then the online resource 120 may invoke a fallback operation during which another agent is selected to generate the response for the corresponding sub-query)
However Bosnjakovic does not clearly disclose:
generating a first contextual value based on the received query; based on the first contextual value and the second contextual value
However MASSIE discloses:
generating a first contextual value based on the received query;
(MASSIE, [0008] generating an input vector based on a user input that indicates a question regarding a first property characteristic of a plurality of property characteristics associated with a knowledge database;…, a first similarity score based on the input vector and the document vector
based on the first contextual value and the second contextual value
(MASSIE [0008] comparing a first similarity score based on the input vector and the document vector to a second similarity score based on the input vector and a response of a plurality of responses of a response database)
Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to incorporate the teaching of Bosnjakovic with the teaching of MASSIE to enable dynamic management and customization of
responses and also to enable questions to be answered based only on relevant information (to reduce hallucinations), (MASSIE, [0065]).
Claims 12 and 13 correspond to claim 1, and are rejected accordingly.
Regarding claim 2, Bosnjakovic in view of MASSIE, discloses all of the features with respect to claim 1 as outlined above. Bosnjakovic does not clearly disclose: generating the first contextual value based on a first data extracted from a knowledgebase, wherein the generative Al software application is configured to generate the response based on data of the knowledgebase.
However MASSIE discloses:
generating the first contextual value based on a first data extracted from a knowledgebase, (MASSIE, [0008] generating an input vector based on a user input that indicates a question regarding a first property characteristic of a plurality of property characteristics associated with a knowledge database;…, a first similarity score based on the input vector and the document vector [0022] providing the input vector to a first machine learning model to identify a document vector within the knowledge database and associated with the first property characteristic based on the input vector; [0020] generating the input vector comprises extracting contextual details from the user input, wherein the input vector is generated based on the contextual details
wherein the generative Al software application is configured to generate the response based on data of the knowledgebase. (MASSIE, [0008] generating an input vector based on a user input that indicates a question regarding a first property characteristic of a plurality of property characteristics associated with a
knowledge database; providing the input vector to a first machine learning model to identify a document vector within the knowledge database and associated with the first
property characteristic based on the input vector; comparing a first similarity score based on the input vector and the document vector to a second similarity score based on the input vector and a response of a plurality of responses of a response database, the response associated with the first property characteristic; providing, based on the comparison of the first similarity score and the second similarity score, the input vector and at least one of (i) the document vector, (ii) the response, or (iii) a combination thereof to a large language model (LLM) to generate response content; and outputting the response content as an answer to the question.)
Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to incorporate the teaching of Bosnjakovic with the teaching of MASSIE to enable dynamic management and customization of responses and also to enable questions to be answered based only on relevant information (to reduce hallucinations), (MASSIE, [0065]).
Claim 14 corresponds to claim 2, and is rejected accordingly.
Regarding claim 3, Bosnjakovic in view of MASSIE discloses all of the features with respect to claim 2 as outlined above. Claim 3 further recites: generating the second contextual value based on a second data extracted from the knowledgebase. (Bosnjakovic, Fig. 1, Database 124, Context 124C; Fig. 5A; [0057]-[0058]; [0048] generate a first vector indicative of the response; [0058] generate a first vector indicative of the response summary; [0035] The database 124 stores user data, product data, service data, and other information associated with the online resource 120…the database 124 is shown to include a user data store 124A, an agent data store 124B, a context data store 124C, and instructions 124D; [0057] the online resource 120 can summarize the responses generated by the selected agents 308 and use the query context information to ensure that the summarized responses are consistent (in context) with their corresponding sub-queries. [0058] compare a respective response summary with the context for the corresponding sub-query to generate a similarity score,… can generate a first vector indicative of the response summary and generate a second vector indicative of the context of the corresponding sub-query, and then generate the similarity score based on a cosine distance between the first vector and the second vector;)
Claim 15 corresponds to claim 3, and is rejected accordingly.
Regarding claim 4, Bosnjakovic in view of MASSIE discloses all of the features with respect to claim 1 as outlined above. Claim 4 further recites: generating in a vector database (Bosnjakovic, Fig. 1; [0035] The database 124 stores user data, product data, service data, and other information associated with the online resource 120…the database 124 is shown to include a user data store 124A, an agent data store 124B, a context data store 124C, and instructions 124D; [0048] The online resource 120 may include a similarity engine 127…the similarity engine 127 can generate a first vector indicative of the response and generate a plurality of second vectors indicative of the corresponding sub-queries; [0058] the online resource 120 can generate a first vector indicative of the response summary and generate a second vector indicative of the context of the corresponding sub-query, and then generate the similarity score based on a cosine distance between the first vector and the second vector.
a first vector corresponding to the first contextual value; (Bosnjakovic, [0058] compare a respective response summary with the context for the corresponding sub-query to generate a similarity score,… can generate a first vector indicative of the response summary and generate a second vector indicative of the context of the corresponding sub-query, and then generate the similarity score based on a cosine distance between the first vector and the second vector.)
generating in the vector database a second vector corresponding to the second contextual value; (Bosnjakovic, [0058] compare a respective response summary with the context for the corresponding sub-query to generate a similarity score,… can generate a first vector indicative of the response summary and generate a second vector indicative of the context of the corresponding sub-query, and then generate the similarity score based on a cosine distance between the first vector and the second vector.)
determining a distance between the first vector and the second vector; (Bosnjakovic, [0058] compare a respective response summary with the context for the corresponding sub-query to generate a similarity score,… can generate a first vector indicative of the response summary and generate a second vector indicative of the context of the corresponding sub-query, and then generate the similarity score based on a cosine distance between the first vector and the second vector.)
and generating the verification score based on the determined distance. (Bosnjakovic, [0058] determine the similarity score for a respective response summary based on Euclidean distances between the first and second vectors.
Claim 16 corresponds to claim 4, and is rejected accordingly.
Regarding claim 11, Bosnjakovic in view of MASSIE discloses all of the features with respect to claim 4 as outlined above. Claim 11 further recites: storing the second vector and the first vector in the vector database; (Bosnjakovic, Fig. 1; [0035] The database 124 stores user data, product data, service data, and other information associated with the online resource 120…the database 124 is shown to include a user data store 124A, an agent data store124B, a context data store 124C, and instructions 124D; [0048] The online resource 120 may include a similarity engine 127…the similarity engine 127 can generate a first vector indicative of the response and generate a plurality of second vectors indicative of the corresponding sub-queries; [0058] the online resource 120 can generate a first vector indicative of the response summary and generate a second vector indicative of the context of the corresponding sub-query, and then generate the similarity score based on a cosine distance between the first vector and the second vector.)
receiving a third vector corresponding to a second query
(Bosnjakovic [0042] determine a context for each of the identified
Queries; [0058] compare a respective response summary with the context for the corresponding sub-query to generate a similarity score,… can generate a first vector indicative of the response summary and generate a second vector indicative of the context of the corresponding sub-query, and then generate the similarity score based on a cosine distance between the first vector and the second vector)
and fourth vector corresponding to a response of the second query; (Bosnjakovic [0058] compare a respective response summary with the context for the corresponding sub-query to generate a similarity score,… can generate a first vector indicative of the response summary and generate a second vector indicative of the context of the corresponding sub-query, and then generate the similarity score based on a cosine distance between the first vector and the second vector.)
determining a distance between the fourth vector and the second vector; (Bosnjakovic [0058] compare a respective response summary with the context for the corresponding sub-query to generate a similarity score,… can generate a first vector indicative of the response summary and generate a second vector indicative of the context of the corresponding sub-query, and then generate the similarity score based on a cosine distance between the first vector and the second vector.)
and providing the response associated with the second vector in response to determining that a distance between the third vector and the fourth vector is below a threshold value. (Bosnjakovic, [0058] then compare the similarity score with a threshold to determine whether the respective response summary is sufficiently similar (in context) with the corresponding subquery to be included as part of the natural language answer…For example, if the similarity score for a respective response summary is greater than the threshold, then the online resource 120 may include the respective response summary in the natural language answer...if the similarity score for a respective response summary is not greater than the threshold, then the online resource 120 may invoke a fallback operation during which another agent is selected to generate the response for the corresponding sub-query)
Claim 23 corresponds to claim 11, and is rejected accordingly.
Claims 5-8 and 17-20 are rejected under 35 U.S.C. 103 as being unpatentable over Bosnjakovic (US 2025/0390516 Al) in view of MASSIE (US 2025/0272581 Al) in view of Mukherjee (US 20240354436 A1) in view of Kane (US 12,639,309 B2)
Regarding claim 5, Bosnjakovic in view of MASSIE discloses all of the features with respect to claim 4 as outlined above. Bosnjakovic in view of MASSIE does not clearly disclose: accessing a data source, the data source including a plurality of textual data; generating a plurality of textual paragraphs based on the plurality of textual data; generating a paragraph vector for each of the plurality of textual paragraphs; and detecting a textual paragraph of the plurality of textual paragraphs utilized by the generative Al software application to generate the received response based on a vector distance between the textual paragraph and the second vector.
However Mukherjee discloses:
accessing a data source, the data source including a plurality of textual data; (Mukherjee, [0103] Next, at (2), the document search module 106 may chunk a set of documents stored in the database module 108 into a plurality of portions/segments of the set of documents. For example, the document search module 106 may chunk documents into a plurality of words, sentences, paragraphs, and/or the like. The text chunks ( e.g., the plurality of portions of the set of documents) may be stored in the ontology 205, or based on the ontology 205.)
generating a plurality of textual paragraphs based on the plurality of textual data;
(Mukherjee, [0103] Next, at (2), the document search module 106 may chunk a set of documents stored in the database module 108 into a plurality of portions/segments of the set of documents. For example, the document search module 106 may chunk documents into a plurality of words, sentences, paragraphs, and/or the like. The text chunks ( e.g., the plurality of portions of the set of documents) may be stored in the ontology 205, or based on the ontology 205.)
generating a paragraph vector for each of the plurality of textual paragraphs;
(Mukherjee, [0104] Then, at (3), the document search module 106 may further vectorize the text chunks to generate a plurality of vectors, where each of the plurality of vectors corresponds to a chunked portion/segment ( e.g., a word, a sentence, a paragraph, or the like) of the set of documents.)
and detecting a textual paragraph of the plurality of textual paragraphs utilized by the generative Al software application to generate the received response based on a vector distance between the textual paragraph and the vector.
(Mukherjee, [0109] At (6), the document search module 106 may execute a similarity search between the query vector generated at (5) and the plurality of vectors generated at (3) to identify one or more documents portions that are more relevant or similar to the natural language user query received at ( 4). As noted above, at (6), the document search module 106 may execute the similarity search using one of the cosine similarity search, approximate nearing neighbor (ANN) algorithms, k nearest neighbors (KNN) method, locality sensitive hashing (LSH), range queries, or any other vector clustering and/or similarity search algorithms; [0104] Then, at (3), the document search module 106 may further vectorize the text chunks to generate a plurality of vectors, where each of the plurality of vectors corresponds to a chunked portion/segment ( e.g., a word, a sentence, a paragraph, or the like) of the set of documents.)
Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to incorporate the teaching of Bosnjakovic in view of MASSIE with the teaching of Mukherjee to enable the system (including the LLM) to more accurately identify portions of the set of documents that are more relevant to the user query, (Mukherjee, [0039]) and also using portions of the set of documents most similar to the user query semantically to generate a prompt to a LLM may enable the system to receive more accurate or desired response from the LLM for the system to responding to the user query, (Mukherjee, [0042])
However Bosnjakovic in view of MASSIE in view of Mukherjee does not clearly disclose:
based on a vector distance between the textual paragraph and the second vector.
However Kane discloses:
based on a vector distance between the textual paragraph and the second vector. (Kane (US 12,639,309 B2), column 25, line 55- response validation
Column 19, lines 42-46, quantifying the influence of each selected block on the generated response may be performed using an embedding similarity-based influence method, wherein text from the selected blocks and the generated response is mapped into a multi-dimensional vector space; column 19, line 48, a numerical representation of each selected block; column 19, lines 60-62, generated response is also mapped into the same embedding space, wherein each portion of the response is represented as a numerical vector); (Kane, provisional application No. 63/567,392, [0037] 6. Response validator, determines the degree to which a response is generated from the corpus. This may be accomplished by embedding a response to a query with the LLM alone (e.g., no corpus), embedding the blocks from which the response was generated, and embedding the response. The response vector embedding may be regarded as a function of the other vector embeddings…also determines the degree to which the response is represented by the set of vectors in the entire corpus. If the validation values are not similar to the blocks or to the entire corpus (e.g., within respective thresholds), then a new response may be generated or the user may receive the message that the response failed to validate in response.)
Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to incorporate the teaching of Bosnjakovic in view of MASSIE in view of Mukherjee with the teaching of Kane to enhance the precision of matching queries with pertinent information blocks within the retrieval-augmented generation (RAG) technique, by ensuring (or increasing the likelihood) that responses are more accurately matched to queries and validating that these responses are indeed reflective of corpus-derived information and to elevate the quality of information delivery in AI-driven query answering services and other models (Kane, [0024]).
Claim 17 corresponds to claim 5, and is rejected accordingly.
Regarding claim 6, Bosnjakovic in view of MASSIE in view of Mukherjee in view of Kane discloses all of the features with respect to claim 5 as outlined above. Bosnjakovic in view of MASSIE in view of Mukherjee does not clearly disclose: generating the second contextual value further based on the detected textual paragraph.
However Kane discloses:
generating the second contextual value further based on the detected textual paragraph. (Kane (US 12,639,309 B2), column 25, line 55- response validation
Column 19, lines 42-46, quantifying the influence of each selected block on the generated response may be performed using an embedding similarity-based influence method, wherein text from the selected blocks and the generated response is mapped into a multi-dimensional vector space; column 19, line 48, a numerical representation of each selected block; column 19, lines 60-62, generated response is also mapped into the same embedding space, wherein each portion of the response is represented as a numerical vector); (Kane, provisional application No. 63/567,392, [0037] 6. Response validator, determines the degree to which a response is generated from the corpus. This may be accomplished by embedding a response to a query with the LLM alone (e.g., no corpus), embedding the blocks from which the response was generated, and embedding the response. The response vector embedding may be regarded as a function of the other vector embeddings…also determines the degree to which the response is represented by the set of vectors in the entire corpus. If the validation values are not similar to the blocks or to the entire corpus (e.g., within respective thresholds), then a new response may be generated or the user may receive the message that the response failed to validate in response.)
Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to incorporate the teaching of Bosnjakovic in view of MASSIE in view of Mukherjee with the teaching of Kane to enhance the precision of matching queries with pertinent information blocks within the retrieval-augmented generation (RAG) technique, by ensuring (or increasing the likelihood) that responses are more accurately matched to queries and validating that these responses are indeed reflective of corpus-derived information and to elevate the quality of information delivery in AI-driven query answering services and other models (Kane, [0024]).
Claim 18 corresponds to claim 6, and is rejected accordingly.
Regarding claim 7, Bosnjakovic in view of MASSIE in view of Mukherjee in view of Kane discloses all of the features with respect to claim 5 as outlined above. Bosnjakovic in view of MASSIE does not clearly disclose: determining a plurality of first distances, each first distance between the first vector and a paragraph vector of a plurality of paragraph vectors; determining a plurality of second distances, each second distance between the second vector and a paragraph vector of the plurality of paragraph vectors; and detecting the textual paragraph based on a first distance of the plurality of first distances which is the shortest and a second distance of the plurality of second distances which is shortest.
However Mukherjee discloses:
determining a plurality of first distances, each first distance between the first vector and a paragraph vector of a plurality of paragraph vectors; (Mukherjee, [0041] Based on the query vector and the plurality of vectors generated from vectorizing portions (e.g., text chunks) of the set of documents permissioned to the user, the system may execute a similarity search between the query vector and the plurality of vectors to identify one or more documents portions that are more relevant or similar to the user query. The system may execute the similarity search using one of the cosine similarity search, approximate nearing neighbor (ANN) algorithms, k nearest neighbors (KNN) method, locality sensitive hashing (LSH), range queries, or any other vector clustering and/or similarity search algorithms… the similarity search may yield similar document portions having a threshold similarity with the first user input. In various examples, the threshold similarity may be adjustable by the system or a user; [0104] Then, at (3), the document search module 106 may further vectorize the text chunks to generate a plurality of vectors, where each of the plurality of vectors corresponds to a chunked portion/segment ( e.g., a word, a sentence, a paragraph, or the like) of the set of documents;)
Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to incorporate the teaching of Bosnjakovic in view of MASSIE with the teaching of Mukherjee to enable the system (including the LLM) to more accurately identify portions of the set of documents that are more relevant to the user query, (Mukherjee, [0039]) and also using portions of the set of documents most similar to the user query semantically to generate a prompt to a LLM may enable the system to receive more accurate or desired response from the LLM for the system to responding to the user query, (Mukherjee, [0042])
Bosnjakovic in view of MASSIE in view of Mukherjee does not clearly disclose:
determining a plurality of second distances, each second distance between the second vector and a paragraph vector of the plurality of paragraph vectors; and detecting the textual paragraph based on a first distance of the plurality of first distances which is the shortest and a second distance of the plurality of second distances which is shortest.
However Kane discloses:
determining a plurality of first distances, each first distance between the first vector and a paragraph vector of a plurality of paragraph vectors; (Kane (US 12,639,309 B2) column 8, lines 16-20- the data set may be indexed by computing embedding vectors in the embedding space for text blocks using hierarchical navigable small world (HNSW) graphs or content-aware chunking strategies to optimize retrieval efficiency; column 13, lines 1-8- determination that the modified prompt embedding vector is within a threshold distance to vectors representing one or more blocks of the data set in the embedding space. A threshold distance, as used in this disclosure, may define a … measure within the embedding space that determines whether a retrieved block is considered sufficiently relevant to the modified prompt embedding vector.; column 8, lines 28-30, maintaining that the modified prompt embedding vector can quickly identify the most relevant blocks in the data set; column 23 lines 15-21, processing the input prompt and converting it into an embedding representation
within a …embedding space… prompt embedding vector to optimize retrieval performance by aligning it with relevant data blocks from the data corpus
250) ; (Kane, provisional application No. 63/567,392 [0032] 2. Vector matching - the query vector may be matched to similar vectors (e.g,. within a threshold distance in the embedding space, measured with distance metrics like cosine distance, Euclidian distance, Minkowski distance, or the like) of blocks from the corpus which represents the information in which it is contained. In some cases, matches may be expedited with techniques like hierarchical navigable small world graphs.)
determining a plurality of second distances, each second distance between the second vector and a paragraph vector of the plurality of paragraph vectors;
(Kane (US 12,639,309 B2), column 25, line 55- response validation
Column 19, lines 42-46, quantifying the influence of each selected block on the generated response may be performed using an embedding similarity-based influence method, wherein text from the selected blocks and the generated response is mapped into a multi-dimensional vector space; column 19, line 48, a numerical representation of each selected block; column 19, lines 60-62, generated response is also mapped into the same embedding space, wherein each portion of the response is represented as a numerical vector); (Kane, provisional application No. 63/567,392 [0037] 6. Response validator, determines the degree to which a response is generated from the corpus. This may be accomplished by embedding a response to a query with the LLM alone (e.g., no corpus), embedding the blocks from which the response was generated, and embedding the response. The response vector embedding may be regarded as a function of the other vector embeddings…also determines the degree to which the response is represented by the set of vectors in the entire corpus. If the validation values are not similar to the blocks or to the entire corpus (e.g., within respective thresholds), then a new response may be generated or the user may receive the message that the response failed to validate in response.)
and detecting the textual paragraph based on a first distance of the plurality of first distances which is the shortest and a second distance of the plurality of second distances which is shortest. (Kane (US 12,639,309 B2), column 14, lines 19-21- closest blocks to the .. representation of the prompt in the embeddings space; column 14, lines 61-64- rank the selected blocks based on their respective proximity measures…blocks with highest relevance… are assigned higher priority for response generation; column 14, lines 36-41, determining proximity between the modified embedding vector and blocks in the data set may include cosine similarity measurements, Euclidean distance computations, Manliattan distance evaluations, Minkowski distance, or kernelized similarity functions; column 15, lines 13-15- maintaining that only blocks with strong alignment to the modified prompt representation are used for generating the response; column 25, line 55- response validation Column 19, lines 42-46, quantifying the influence of each selected block on the generated response may be performed using an embedding similarity-based influence method, wherein text from the selected blocks and the generated response is mapped into a multi-dimensional vector space; column 19, line 48, a numerical representation of each selected block; column 19, lines 60-62, generated response is also mapped into the same embedding space, wherein each portion of the response is represented as a numerical vector); (Kane, provisional application No. 63/567,392, [0033] 3. Response generation - the block or blocks from the corpus that are the most similar (e.g., the closest in vector space, a closes number, or those within a threshold distance) to the query vector may be used in an LLM prompt to create a response. [0025] These weights (and biases in some cases) may be used for assessing the similarity between a query and a text block from the corpus…quantifies the degree to which a response is derived from corpus blocks)
Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to incorporate the teaching of Bosnjakovic in view of MASSIE in view of Mukherjee with the teaching of Kane to enhance the precision of matching queries with pertinent information blocks within the retrieval-augmented generation (RAG) technique, by ensuring (or increasing the likelihood) that responses are more accurately matched to queries and validating that these responses are indeed reflective of corpus-derived information and to elevate the quality of information delivery in AI-driven query answering services and other models (Kane, [0024]).
Claim 19 corresponds to claim 7, and is rejected accordingly.
Regarding claim 8, Bosnjakovic in view of MASSIE in view of Mukherjee in view of Kane discloses all of the features with respect to claim 5 as outlined above. Bosnjakovic does not clearly disclose:
detecting the textual paragraph by providing a prompt to a language model including the received query and the received response.
However MASSIE discloses:
by providing a prompt to a language model including the received query and the received response. (MASSIE, [0087] information provided to the LLM 410 (such as the input vector 414 and one of the document vector 414 and the predetermined response 416) may be provided as a prompt to the LLM 410… prompts provided to the LLM 410 may exclude previous inputs received from a user (such as earlier portions of a conversation in which the user input 402 is received; [0084] the predetermined response 416 may be determined by a machine learning model 408)
Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to incorporate the teaching of Bosnjakovic with the teaching of MASSIE to enable dynamic management and customization of responses and also to enable questions to be answered based only on relevant information (to reduce hallucinations), (MASSIE, [0065]).
However Mukherjee discloses:
detecting the textual paragraph
(Mukherjee, [0109] At (6), the document search module 106 may execute a similarity search between the query vector generated at (5) and the plurality of vectors generated at (3) to identify one or more documents portions that are more relevant or similar to the natural language user query received at ( 4). As noted above, at (6), the document search module 106 may execute the similarity search using one of the cosine similarity search, approximate nearing neighbor (ANN) algorithms, k nearest neighbors (KNN) method, locality sensitive hashing (LSH), range queries, or any other vector clustering and/or similarity search algorithms; [0104] Then, at (3), the document search module 106 may further vectorize the text chunks to generate a plurality of vectors, where each of the plurality of vectors corresponds to a chunked portion/segment ( e.g., a word, a sentence, a paragraph, or the like) of the set of documents; [0041] the similarity search may yield similar document portions having a threshold similarity with the first user input)
Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to incorporate the teaching of Bosnjakovic in view of MASSIE with the teaching of Mukherjee to enable the system (including the LLM) to more accurately identify portions of the set of documents that are more
relevant to the user query, (Mukherjee, [0039]) and also using portions of the set of documents most similar to the user query semantically to generate a prompt to a LLM may enable the system to receive more accurate or desired response from the LLM for the system to responding to the user query, (Mukherjee, [0042])
Claim 20 corresponds to claim 8, and is rejected accordingly.
Claims 9-10 and 21-22 are rejected under 35 U.S.C. 103 as being unpatentable over Bosnjakovic (US 2025/0390516 Al) in view of MASSIE (US 2025/0272581 Al) in view of Mukherjee (US 20240354436 A1)
Regarding claim 9, Bosnjakovic in view of MASSIE discloses all of the features with respect to claim 4 as outlined above. Bosnjakovic in view of MASSIE does not clearly disclose: accessing a plurality of data sources, each data source including textual data; generating for each textual data a plurality of textual paragraphs; and generating for each text paragraph of the plurality of text paragraphs a plurality of sentences.
However Mukherjee discloses:
accessing a plurality of data sources, each data source including textual data; generating for each textual data a plurality of textual paragraphs; and generating for each text paragraph of the plurality of text paragraphs a plurality of sentences. (Mukherjee, [0103] Next, at (2), the document search module 106 may chunk a set of documents stored in the database module 108 into a plurality of portions/segments of the set of documents. For example, the document search module 106 may chunk documents into a plurality of words, sentences, paragraphs, and/or the like. The text chunks ( e.g., the plurality of portions of the set of documents) may be stored in the ontology 205, or based on the ontology 205.)
Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to incorporate the teaching of Bosnjakovic in view of MASSIE with the teaching of Mukherjee to enable the system (including the LLM) to more accurately identify portions of the set of documents that are more relevant to the user query, (Mukherjee, [0039]) and also using portions of the set of documents most similar to the user query semantically to generate a prompt to a LLM may enable the system to receive more accurate or desired response from the LLM for the system to responding to the user query, (Mukherjee, [0042])
Claim 21 corresponds to claim 9, and is rejected accordingly.
Regarding claim 10, Bosnjakovic in view of MASSIE in view of Mukherjee discloses all of the features with respect to claim 9 as outlined above. Bosnjakovic in view of MASSIE does not clearly disclose: generating each text paragraph of the plurality of paragraphs based on metadata associated with the textual data.
However Mukherjee discloses:
generating each text paragraph of the plurality of paragraphs based on metadata associated with the textual data. (Mukherjee, [0103] The text chunks ( e.g., the plurality of portions of the set of documents) may be stored in the ontology 205, or based on the ontology 205. The document search module 106 may select the granularity (e.g., words, sentences, paragraphs, and/or the like) for chunking documents based on various criterion, such as a size of the set of documents, a type of the set of documents, a type of similarity search as described herein, a user feedback as described herein, and/or the like.)
Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to incorporate the teaching of Bosnjakovic in view of MASSIE with the teaching of Mukherjee to enable the system (including the LLM) to more accurately identify portions of the set of documents that are more relevant to the user query, (Mukherjee, [0039]) and also using portions of the set of documents most similar to the user query semantically to generate a prompt to a LLM may enable the system to receive more accurate or desired response from the LLM for the system to responding to the user query, (Mukherjee, [0042])
Claim 22 corresponds to claim 10, and is rejected accordingly.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Faezeh Forouharnejad whose telephone number is (571)270-7416. The examiner can normally be reached on Mondays, Wednesdays and Thursdays.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Shah Sanjiv can be reached on (571)272-4098. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from Patent Center and the Private Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from Patent Center or Private PAIR. Status information for unpublished applications is available through Patent Center and Private PAIR to authorized users only. Should you have questions about access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free)
/F.F. /
Examiner, Art Unit 2166
/SANJIV SHAH/ Supervisory Patent Examiner, Art Unit 2166
Prosecution Insights