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
Claims 1 and 11 have been amended. Claims 24-25 have been added. Claims 20-22 have been canceled. Claims 1-2, 4-12, 14-19, and 23-25 are pending in the application.
Response to Arguments
Claim Rejections - 35 USC § 101
Rejection of claims 1-2, 4- 12, and 14-23 have been withdrawn in response to applicant's amendments and arguments.
Claim Rejections - 35 USC § 112
Applicant submits that the specification at paragraph [0204] and [0214] provides the scope of the invention such that one of ordinary skill in the art would know what was meant, and that the subject matter of the invention has been described and is supported in such a way as to reasonably convey to one skilled in the relevant art that the inventors, at the time the application was filed, had possession of the claimed invention. Particularly, the claim recitations of "wherein the context embeddings comprising floating point numbers inserted into the one or more vectors in association with the text" is expressly supported by "The content embeddings can include floating point numbers generated by running the text through a model or a transformer. The content embeddings can be inserted into the vectors." Of paragraph [0214].
Examiner respectfully submits that in the specification, floating point numbers are described only in connection with content embeddings, Not context embeddings. The specification doses not disclose “wherein the context embeddings comprising floating point numbers inserted into the one or more vectors in association with the text”. The specification in paragraph [0214] discloses “The content embeddings can include floating point numbers generated by running the text through a model or a transformer. The content embeddings can be inserted into the vectors”.
Although both content embedding and context embedding may comprise numerical values, an embedding is a representation of underlying information.
The specification describes both content embeddings and context embeddings but with regards to “floating point numbers” feature, only discloses “The content embeddings can include floating point numbers”. The claimed limitation requires the context embeddings to comprise the floating numbers inserted into the one or more vectors in association with the text. Therefore, the claims 1-2, 4-12, 14-19, and 23-25 remain rejected.
Claim Rejections - 35 USC § 103
In response to the amendments and arguments, Examiner relies on a new combination of references.
Claim Objections
Claims 25 is objected to because of the following informality:
Claim 25 recites, in the last line, “the advanced agentic searchi”. Examiner suggests that “the advanced agentic searchi” be replaced with “the advanced agentic searching”.
Appropriate correction is required.
Claim Rejections - 35 USC § 112
The following is a quotation of the first paragraph of 35 U.S.C. 112(a):
(a) IN GENERAL.—The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor or joint inventor of carrying out the invention.
The following is a quotation of the first paragraph of pre-AIA 35 U.S.C. 112:
The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor of carrying out his invention.
Claims 1-2, 4-12, 14-19, and 23-25 are rejected under 35 U.S.C. 112(a) or 35 U.S.C. 112 (pre-AIA ), first paragraph, as failing to comply with the written description requirement. The claim(s) contains subject matter which was not described in the specification in such a way as to reasonably convey to one skilled in the relevant art that the inventor or a joint inventor, or for pre-AIA the inventor(s), at the time the application was filed, had possession of the claimed invention.
Claims 1, and 11 recite “wherein the context embeddings comprising floating point numbers inserted into the one or more vectors in association with the text”. The specification shows the following about “floating point numbers”: “[0204] The content embeddings can take the form of a long vector of floating point numbers that is produced by running the text through a model or a transformer. By using a vectors, the one or more Al agents can utilize the content embeddings to quickly access the text, which saves processing time and improves processing accuracy. [0214] The content embeddings can include floating point numbers generated by running the text through a model or a transformer. The content embeddings can be inserted into the vectors.”.
In the specification, floating point numbers are described only in connection with content embeddings. The specification doses not disclose “wherein the context embeddings comprising floating point numbers inserted into the one or more vectors in association with the text”.
Dependent claims are also rejected for inheriting the deficiencies of the base claim.
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 1-2, 4- 12, 14-19 and 23-24 are rejected under 35 U.S.C. 103 as being unpatentable over Seabra et al. (“Dynamic Multi-Agent Orchestration and Retrieval for Multi-Source Question-Answer Systems using Large Language Models”, cited in IDS, hereinafter Seabra ) in view of JAIN (US 20250181623 A1) in view of LONG (US 2026/0081881 Al)
Regarding claim 1, Seabra discloses: An agentic searching method implemented by one or more artificial intelligence (AI) agents that utilize content embeddings to access a semantic storage to save processing time and improve processing accuracy of a complex query, the one or more Al agents stored as program code on a non-transitory computer readable medium and executable by one or more processors, (Seabra, page 2, 1st paragraph- a dynamic multi-agent orchestration and retrieval methodology aimed at improving the accuracy of multi-source Q&A systems using Large Language Models (corresponding to “artificial intelligence (AI)”; section 2.5 Agents- These agents leverage recent advancements in AI)
the agentic searching method comprising: performing, at runtime by the one or more Al agents, advanced agentic searching of (Seabra, section 2.5 Agents- These agents leverage recent advancements in AI; page 2, 1st paragraph- a dynamic multi-agent orchestration and retrieval methodology aimed at improving the accuracy of multi-source Q&A systems using Large Language Models. By combining advanced retrieval-augmented generation (RAG), text-to-SQL techniques, and dynamic prompt engineering, we enable the system to handle complex queries … an agent-based architecture that dynamically orchestrates different retrieval strategies based on the nature of the user query, ensuring optimal data retrieval from multiple sources; page 3, section 2.2, e.g. 2nd paragraph- The RAG framework…operates by embedding both the user's query and chunks of external information into high-dimensional vector spaces. These embeddings allow the system to compare and retrieve the most semantically relevant data from a vectorstore; page 8, section 3.1, line 6-The vector store allows efficient querying of vectors; page 8, section 3.1, line 5-These embeddings (vectors) are stored in a vectorstore; page 12, section 3.2, 1st pargraph- as it enabled direct integration of database information into our existing RAG framework, ensuring that queries could retrieve relevant data without needing to connect to the database during runtime; page 13, section 3.3. Agents, e.g. the system dynamically routes each query to the most suitable processing pathway, ensuring that user questions are handled with precision and contextual relevance. In our architecture, the Router Agent serves as the primary decision-maker, evaluating each incoming query and directing it to the appropriate agent based on predefined criteria )
a semantic storage structurally comprising one or more vectors containing a text and context embeddings (Seabra, page 8, section 3.1, Applying RAG- (1) reading the textual content of the PDF documents into manageable (chunks), which are then ( 2) transformed into high-dimensional vectors (embedding). The text in vector format captures the semantic properties of the text (corresponding to “context”), a format that can have 1536 dimensions or more. These embeddings (vectors) are stored in a vectorstore (corresponding to “sematic storage”) (3), a database specialized in high-dimensional vectors… (5), which are then transparently integrated into the LLM context)
in response to the complex query by computing search vectors from the complex query and using the search vectors to search the context embedding of the one or more vectors, (page 3, section 2.2., e.g. The RAG framework, operates by embedding both the user's query (corresponding to “search vectors”) and chunks of external information into high-dimensional vector spaces. These embeddings allow the system to compare and retrieve the most semantically relevant data from a vectorstore - a database optimized for high-dimensional vectors. Once retrieved, this relevant data is used as additional input to the LLM, ensuring that the generated answer is informed by the latest and most pertinent information available.)
outputting a context grounding resulting from the searching of the semantic storage and the context embeddings to an Al model with the complex query to enhance an accuracy of a response of the Al model; (Seabra, page 8, section 3.1, Once the similarity metric is established, the query is embedded in the same vector space (4); this allows a direct comparison between the vectorized query and the vectors of the stored chunks, retrieving the most similar chunks (5), which are then transparently integrated into the LLM (corresponding to “AI model”)context to generate a prompt (6). The prompt is then composed of the question, the texts retrieved from the vectorstore, the specific instructions and, optionally, the chat history, all sent to the LLM (corresponding to “AI model”) which generates the final response; page 6- For example, by using instructions like "Do not use prior knowledge", prompt engineering can restrict the LLM's answer to specific sources, thereby reducing the risk of factual hallucinations and ensuring that responses are grounded in the desired information; section 2.2 - RAG addresses this by retrieving relevant documents or data from external sources, such as databases or document repositories, and feeding this information to the LLM as context for generating responses…The chunking strategy used in RAG is critical to its success, as it determines how documents are divided into smaller pieces for embedding and retrieval)
and receiving the response to the complex query from the Al model. (Seabra, section 3.1- The prompt is then composed of the question, the texts retrieved from the vectorstore, the specific instructions and, optionally, the chat history, all sent to the LLM (corresponding to “Al model”) which generates the final response)
However, Seabra does not disclose:
wherein the semantic storage structurally indexes the one or more vectors and the context embeddings to reduce processing time of the advanced agentic searching at runtime, wherein the context embeddings comprising floating point numbers inserted into the one or more vectors in association with the text, wherein the floating point numbers of the content embeddings are generated from the text through a model or a transformer to decrease a complexity of the text, which reduces a processing or lookup time during the advanced agentic searching of the complex query;
However, JAIN discloses:
wherein the context embeddings comprising floating point numbers inserted into the one or more vectors in association with the text, (JAIN, [0037] OpenAI's embeddings model is a vector of floating-point numbers that represents the “meaning” of text. The distance between two vectors serves as a gauge of their degree of relatedness. Smaller distances indicate a higher degree of relatedness, whereas larger distances signify lower relatedness; [0036] Embeddings (122) are numerical representations of concepts (data) converted to number sequences Embeddings are mathematical portrayals of words or phrases, enabling the comparison of distinct blocks of text; [0035] Vector stores (120) are databases that store embeddings for different phrases or words. By using a vector store, developers may quickly access pre-computed embeddings, which may save time and improve the accuracy of the model's responses.)
wherein the floating point numbers of the content embeddings are generated from the text through a model or a transformer to decrease a complexity of the text, which reduces a processing or lookup time during the advanced agentic searching of the complex query; (JAIN, [0037] OpenAI's embeddings model is a vector of floating-point numbers that represents the “meaning” of text. The distance between two vectors serves as a gauge of their degree of relatedness. Smaller distances indicate a higher degree of relatedness, whereas larger distances signify lower relatedness; [0036] Embeddings (122) are numerical representations of concepts (data) converted to number sequences Embeddings are mathematical portrayals of words or phrases, enabling the comparison of distinct blocks of text; [0035] Vector stores (120) are databases that store embeddings for different phrases or words. By using a vector store, developers may quickly access pre-computed embeddings, which may save time and improve the accuracy of the model's responses.)
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 Seabra with the teaching of JAIN to quickly access pre-computed embeddings, which may save time and improve the accuracy of the model's responses, (JAIN, [0035]) and also to provide faster query service and responses, (JAIN, [0034]).
However, Seabra in view of JAIN does not disclose:
wherein the semantic storage structurally indexes the one or more vectors and the context embeddings to reduce processing time of the advanced agentic searching at runtime,
However, LONG discloses:
wherein the semantic storage structurally indexes the one or more vectors and the context embeddings to reduce processing time of the advanced agentic searching at runtime, (LONG [0028] the vector index may be used to store the data chunks from 204 and/or the calculated embedding from 206. A vector index is a data structure to efficiently store and retrieve high-dimensional vector data (e.g., the data chunk and/or its embedding), enabling fast similarity searches and nearest neighbor queries. This vector indexing technique may involve neatly arranging the high-dimensional vectors in a searchable and efficient manner. [0024] In the runtime stage 250, response generation system 100 may receive a user query (e.g., query context and/ or instruction). Response generation system 100 may also calculate embedding for the user query, and then perform a similarity search to retrieve the relevant texts and/or documents within database 170 that are semantically close to the us;)
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 Seabra in view of JAIN with the teaching of LONG to allow quick and accurate similarity searches and pattern identification especially for searching large and complex datasets and neatly arranging the vectors in a searchable and efficient manner, (LONG, [0028]) and also allowing for efficient and accurate semantic and similarity search to be performed, (LONG, [0027]).
Claim 11 corresponds to claim 1, and is rejected accordingly.
Regarding claim 2, Seabra in view of JAIN in view of LONG discloses all of the features with respect to claim 1 as outlined above. Claim 2 further recites: wherein the advanced agentic searching retrieves information that is grounded to the context of the complex query. (Seabra, page 3, section 2.2, e.g. 2nd paragraph- The RAG framework…operates by embedding both the user's query and chunks of external information into high-dimensional vector spaces. These embeddings allow the system to compare and retrieve the most semantically relevant data (corresponding to “information that is grounded to the context of the complex query”) from a vectorstore; page 8, section 3.1, Once the similarity metric is established, the query is embedded in the same vector space (4); this allows a direct comparison between the vectorized query and the vectors of the stored chunks, retrieving the most similar chunks (5), which are then transparently integrated into the LLM context to generate a prompt (6). The prompt is then composed of the question, the texts retrieved from the vectorstore, the specific instructions and, optionally, the chat history, all sent to the LLM which generates the final response; page 6, 1st paragraph- ensuring that responses are grounded in the desired information;)
Claim 12 corresponds to claim 2, and is rejected accordingly.
Regarding claim 4, Seabra in view of JAIN in view of LONG discloses all of the features with respect to claim 1 as outlined above. Claim 4 further recites: wherein the advanced agentic searching comprises a semantic search and retrieval operation matching the complex query to the context embeddings of the semantic storage. (Seabra , page 3, section 2.2, e.g. 2nd paragraph- The RAG framework…operates by embedding both the user's query and chunks of external information into high-dimensional vector spaces. These embeddings allow the system to compare and retrieve the most semantically relevant data from a vectorstore (corresponding to “a semantic search and retrieval operation matching the complex query to the context embeddings of the sematic storage”) ; page 8, section 3.1, Applying RAG- (1) reading the textual content of the PDF documents into manageable (chunks), which are then ( 2) transformed into high-dimensional vectors (embedding). The text in vector format captures the semantic properties of the text (corresponding to “context”), a format that can have 1536 dimensions or more. These embeddings (vectors) are stored in a vectorstore (corresponding to “ semantic storage”)(3), a database specialized in high-dimensional vectors… (5), which are then transparently integrated into the LLM context)
Claim 14 corresponds to claim 4, and is rejected accordingly.
Regarding claim 5, Seabra in view of JAIN in view of LONG discloses all of the features with respect to claim 1 as outlined above. Claim 5 further recites: wherein the advanced agentic searching comprises a hybrid search operation comprising a vector computing with keyword search. (Seabra, page 15, the last paragraph- By integrating both the vectorstore and structured database, the Backend Agents can provide comprehensive answers to user queries, dynamically choosing the most appropriate data source based on the type of question. This hybrid approach ensures that the system can handle both semantically complex queries and direct database queries, offering flexible and accurate responses; page 13, section 3.3- The Router Agent uses regular expressions to identify keywords, patterns, or structures within the query. If the query is specific to a clause within a contract, the Router Agent recognizes this pattern and assigns the query to the RAG Agent. The RAG Agent is optimized for handling unstructured text data, retrieving relevant text chunks from the vectorstore. By focusing on textual similarity, the RAG Agent retrieves semantically aligned information and generates responses.)
Claim 15 corresponds to claim 5, and is rejected accordingly.
Regarding claim 6, Seabra in view of JAIN in view of LONG discloses all of the features with respect to claim 1 as outlined above. Claim 6 further recites: wherein the advanced agentic searching comprises a broad search (Seabra, page 3, section 2.2, second paragraph- These embeddings allow the system to compare and retrieve the most semantically relevant data from a vectorstore ... Once retrieved, this relevant data is used as additional input to the LLM, ensuring that the generated answer is informed by the latest and most pertinent information available; page 3, section 2.2, last paragraph- RAG ensures that only the most relevant sections are retrieved and fed into the LLM, preventing information overload and improving the precision of the answer.) (Note: Examiner interprets that “semantic search” is corresponding to “a broad search”)
with re-ranking operation. (Seabra, page 10, 1st paragraph- For instance, when querying, "Who is the contract manager of contract 123/2024?" the system first filters for chunks that belong to contract number 123/2024 and clauses related to the contract manager. Once these chunks are filtered, a similarity calculation is applied to identify the most relevant text segments (corresponding to re-ranking operation”), which are then sent to the LLM to generate the final response; page 8, section 3.1 efficient querying of vectors through their similarities, using the distance for comparison (whether A1anhatan, Euclidean or cosine). Once the similarity metric is established, the query is embedded in the same vector space (4); this allows a direct comparison between the vectorized query and the vectors of the stored chunks, retrieving the most similar chunks (5), which are then transparently integrated into the LLM context to generate a prompt; page 12, second paragraph- In our work, we chose cosine similarity for its ability to prioritize semantic alignment between query embeddings and document embeddings;)
Claim 16 corresponds to claim 6, and is rejected accordingly.
Regarding claim 7, Seabra in view of JAIN in view of LONG discloses all of the features with respect to claim 1 as outlined above. Claim 7 further recites: wherein the one or more Al agents implements tethering of the response based on dynamic or direct inputs. (Seabra, page 6, section 2.5, e.g. second paragraph- In an agent-based framework, specialized agents can be designed to handle specific tasks, each tailored to different aspects of information retrieval. For example, a Router Agent can serve as the system's primary decision-maker, analyzing each query upon receipt (corresponding to “dynamic input”) and deciding on the optimal retrieval strategy… These agents leverage recent advancements in AI, such as RAG and Text-to-SQL, to perform more complex and contextually aware tasks; page 3, section 2.2- RAG addresses this by retrieving relevant documents or data from external sources, such as databases or document repositories, and feeding this information to the LLM as context for generating responses; page 14, section 3.4, Dynamic Prompt Engineering, 1st paragraph- we use Prompt Engineering (corresponding to “tethering”) to enhance the accuracy of generated answers, guiding the LLM's behavior to ensure responses are contextually relevant and tailored to the user's needs; page 14, section 3.4, 4th paragraph - If a user asks the RAG Agent, "What are the responsibilities of the contract manager for contract 123/2024?", the prompt is constructed as follows: "Retrieve relevant sections from contract 123/2024 that detail the role and responsibilities of the contract manager. Use information from clauses specifying contract management tasks." This tailored prompt focuses the LLM on extracting specific responsibilities, enhancing relevance and accuracy.)
Claim 17 corresponds to claim 7, and is rejected accordingly.
Regarding claim 8, Seabra in view of JAIN in view of LONG discloses all of the features with respect to claim 1 as outlined above. Claim 8 further recites: wherein the one or more Al agents implements automatic tethering. (Seabra, page 18- The use of backend agents, particularly the Router Agent, allowed for a flexible and adaptive workflow where queries are dynamically routed to the appropriate processing module (corresponding to “automatic tethering”)-whether that be the RAG agent for text-based retrieval or the SQL agent for direct database queries; page 14, section 3.4, 4th paragraph - If a user asks the RAG Agent, "What are the responsibilities of the contract manager for contract 123/2024?", the prompt is constructed as follows: "Retrieve relevant sections from contract 123/2024 that detail the role and responsibilities of the contract manager. Use information from. clauses specifying contract management tasks." This tailored prompt focuses the LLM on extracting specific responsibilities, enhancing relevance and accuracy; page 17, second paragraph- Additionally, users appreciated the system's capacity to automatically generate visual summaries (corresponding to “automatic tethering”) through its Plotly agent ,when a table of values was included in the response. This feature was positively received, as it not only provides immediate visual insights but also supports users in preparing professional presentations. By integrating dynamic graph generation directly into the response process,…)
Claim 18 corresponds to claim 8, and is rejected accordingly.
Regarding claim 9, Seabra in view of JAIN in view of LONG discloses all of the features with respect to claim 1 as outlined above. Claim 9 further recites: wherein the context grounding comprise relevant information from unique industry terminology and complex document structures (Seabra, page 10, section Embeddings models, e.g. line 21- to train an embedding model on a domain-specific corpus. This can help the model better capture the unique terminology and context of that field; page 18- The use of backend agents, particularly the Router Agent, allowed for a flexible and adaptive workflow where queries are dynamically routed to the appropriate processing module-whether that be the RAG agent for text-based retrieval or the SQL agent for direct database queries; page 14, section 3.4, 4th paragraph - If a user asks the RAG Agent, "What are the responsibilities of the contract manager for contract 123/2024?" (corresponding to “unique industry terminology and complex document structures “) , the prompt is constructed as follows: "Retrieve relevant sections from contract 123/2024 that detail the role and responsibilities of the contract manager. Use information from. clauses specifying contract management tasks." This tailored prompt focuses the LLM on extracting specific responsibilities, enhancing relevance and accuracy; )
not available to the Al model. (Seabra, page 3, section 2.2 Retrieval-Augmented Generation (RAG) Retrieval-Augmented Generation (RAG) is an advanced technique designed to enhance the performance of LLMs by incorporating external data into the generation process, enabling an extension of knowledge by accessing information that is not part of the LLMs internal knowledge base. While LLMs excel at generating text based on their training, they are limited by the information they contain, which becomes outdated or incomplete over time. RAG addresses this by retrieving relevant documents or data from external sources, such as databases or document repositories, and feeding this information to the LLM as context for generating responses. This ensures that the answers provided are up-to-date and rooted in real-world;)
Claim 19 corresponds to claim 9, and is rejected accordingly.
Regarding claim 10, Seabra in view of JAIN in view of LONG discloses all of the features with respect to claim 1 as outlined above. Claim 10 further recites: wherein the Al model comprises a large language model. (Seabra, section 2.1, Large Language Models- The advent of LLMs such as GPT has significantly advanced the field of Q&A systems, providing an intuitive interface for retrieving information from diverse data sources.)
Regarding claim 23, Seabra in view of JAIN in view of LONG discloses all of the features with respect to claim 1 as outlined above. Claim 23 further recites: wherein the advanced agentic searching comprises a semantic search and retrieval operation matching the complex query to the context embeddings of the semantic storage, (Seabra , page 3, section 2.2, e.g. 2nd paragraph- The RAG framework…operates by embedding both the user's query and chunks of external information into high-dimensional vector spaces. These embeddings allow the system to compare and retrieve the most semantically relevant data from a vectorstore (corresponding to “a semantic search and retrieval operation matching the complex query to the context embeddings of the sematic storage”) ; page 8, section 3.1, Applying RAG- (1) reading the textual content of the PDF documents into manageable (chunks), which are then ( 2) transformed into high-dimensional vectors (embedding). The text in vector format captures the semantic properties of the text (corresponding to “context”), a format that can have 1536 dimensions or more. These embeddings (vectors) are stored in a vectorstore (corresponding to “ semantic storage”)(3), a database specialized in high-dimensional vectors… (5), which are then transparently integrated into the LLM context)
a hybrid search operation comprising a vector computing with keyword search, (Seabra, page 15, the last paragraph- By integrating both the vectorstore and structured database, the Backend Agents can provide comprehensive answers to user queries, dynamically choosing the most appropriate data source based on the type of question. This hybrid approach ensures that the system can handle both semantically complex queries and direct database queries, offering flexible and accurate responses; page 13, section 3.3- The Router Agent uses regular expressions to identify keywords, patterns, or structures within the query. If the query is specific to a clause within a contract, the Router Agent recognizes this pattern and assigns the query to the RAG Agent. The RAG Agent is optimized for handling unstructured text data, retrieving relevant text chunks from the vectorstore. By focusing on textual similarity, the RAG Agent retrieves semantically aligned information and generates responses.)
and a broad search (Seabra, page 3, section 2.2, second paragraph- These embeddings allow the system to compare and retrieve the most semantically relevant data from a vectorstore ... Once retrieved, this relevant data is used as additional input to the LLM, ensuring that the generated answer is informed by the latest and most pertinent information available; page 3, section 2.2, last paragraph- RAG ensures that only the most relevant sections are retrieved and fed into the LLM, preventing information overload and improving the precision of the answer.) (Note: Examiner interprets that “semantic search” is corresponding to “a broad search”)
with re-ranking operation.
(Seabra, page 10, 1st paragraph- For instance, when querying, "Who is the contract manager of contract 123/2024?" the system first filters for chunks that belong to contract number 123/2024 and clauses related to the contract manager. Once these chunks are filtered, a similarity calculation is applied to identify the most relevant text segments (corresponding to re-ranking operation”), which are then sent to the LLM to generate the final response; page 8, section 3.1 efficient querying of vectors through their similarities, using the distance for comparison (whether A1anhatan, Euclidean or cosine). Once the similarity metric is established, the query is embedded in the same vector space (4); this allows a direct comparison between the vectorized query and the vectors of the stored chunks, retrieving the most similar chunks (5), which are then transparently integrated into the LLM context to generate a prompt; page 12, second paragraph- In our work, we chose cosine similarity for its ability to prioritize semantic alignment between query embeddings and document embeddings;)
Regarding claim 24, Seabra in view of JAIN in view of LONG discloses all of the features with respect to claim 1 as outlined above. Seabra in view of JAIN does not clearly disclose: wherein the semantic storage structurally indexes the one or more vectors and the context embeddings by pre- calculating distances between the context embeddings of the one or more vectors and organizing similar vectors close to one another in clusters or a graph according to an approximate nearest neighbor approach, such that the search vectors computed from the complex query are searched against a reduced set of semantically proximate vectors at runtime.
However, LONG discloses:
wherein the semantic storage structurally indexes the one or more vectors and the context embeddings by pre- calculating distances between the context embeddings of the one or more vectors (LONG, [0028] the vector index may be used to store the data chunks from 204 and/or the calculated embedding from 206. A vector index is a data structure to efficiently store and retrieve high-dimensional vector data (e.g., the data chunk and/or its embedding), enabling fast similarity searches and nearest neighbor queries. [0024] In the runtime stage 250, response generation system 100 may receive a user query (e.g., query context and/ or instruction). Response generation system 100 may also calculate embedding for the user query, and then perform a similarity search to retrieve the relevant texts and/or documents within database 170 that are semantically close to the user query, and/or its calculated embedding.)
and organizing similar vectors close to one another in clusters or a graph according to an approximate nearest neighbor approach, such that the search vectors computed from the complex query are searched against a reduced set of semantically proximate vectors at runtime. (LONG , [0028] This vector indexing technique may involve neatly arranging the high-dimensional vectors in a searchable and efficient manner. This arrangement may be done in a way that similar vectors and/or embedding are grouped together; [0033] response generation system 100 may apply vector search to find similar data using ANN algorithms. When a query point is provided, the ANN algorithm may use an index to quickly identify a set of candidate points that are likely to be close to the query point. This way, when querying the vector database 210 to find the nearest neighbors of a query vector, instead of computing distances between the query vector and all vectors in the vector database 210, response generation system 100 may only compute distances between the query vector and the small number of candidate vectors around the query vector; [0013] Algorithms like nearest neighbor and approximate nearest neighbor (ANN) may be leveraged by the response generation system as efficient methods to process and rank large volumes of texts and/or documents in the database for AI data grounding of search queries.)
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 Seabra in view of JAIN with the teaching of LONG to allow quick and accurate similarity searches and pattern identification especially for searching large and complex datasets and neatly arranging the vectors in a searchable and efficient manner, (LONG, [0028]) and also allowing for efficient and accurate semantic and similarity search to be performed, (LONG, [0027]).
Claim 25 is rejected under 35 U.S.C. 103 as being unpatentable over Seabra et al. (“Dynamic Multi-Agent Orchestration and Retrieval for Multi-Source Question-Answer Systems using Large Language Models”, cited in IDS, hereinafter Seabra ) in view of JAIN (US 20250181623 A1) in view of LONG (US 2026/0081881 Al) in view of Hemington (US12158906B2)
Regarding claim 25, Seabra in view of JAIN in view of LONG discloses all of the features with respect to claim 1 as outlined above. Claim 25 further recites:
wherein the one or more Al agents implement automatic tethering by saving the context grounding or one or more results of the advanced agentic searching in an agentic memory, (Seabra, page 18- The use of backend agents, particularly the Router Agent, allowed for a flexible and adaptive workflow where queries are dynamically routed to the appropriate processing module (corresponding to “automatic tethering”)-whether that be the RAG agent for text-based retrieval or the SQL agent for direct database queries; page 14, section 3.4, 4th paragraph - If a user asks the RAG Agent, "What are the responsibilities of the contract manager for contract 123/2024?", the prompt is constructed as follows: "Retrieve relevant sections from contract 123/2024 that detail the role and responsibilities of the contract manager. Use information from. clauses specifying contract management tasks." This tailored prompt focuses the LLM on extracting specific responsibilities, enhancing relevance and accuracy; page 17, second paragraph- Additionally, users appreciated the system's capacity to automatically generate visual summaries (corresponding to “automatic tethering”) through its Plotly agent ,when a table of values was included in the response. This feature was positively received, as it not only provides immediate visual insights but also supports users in preparing professional presentations. By integrating dynamic graph generation directly into the response process,…)
Seabra in view of JAIN in view of LONG does not clearly disclose:
and for a subsequent complex query that is semantically similar to the complex query, semantically searching the agentic memory to retrieve the saved context grounding or the one or more results to avoid re-performing at least a portion of the advanced agentic searchi
However, Hemington discloses:
and for a subsequent complex query that is semantically similar to the complex query, semantically searching the agentic memory to retrieve the saved context grounding or the one or more results to avoid re-performing at least a portion of the advanced agentic searchi (Hemington, column 4, line17- to reduce the time taken to find a solution to an incoming query, it may be desirable for an LLM to compare the incoming query to previous queries that contained the same question. The solutions corresponding to such previous queries may be known and stored in association with the queries; column 4, line 44- the system may generate responses to an incoming query by matching the query to a particular one of the clusters and obtaining response messages based on data associated with the matching cluster.)
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 Seabra in view of JAIN in view of LONG with the teaching of Hemington to choose the most similar queries as part of generating a response to the incoming query to reduce the time taken to find a solution to an incoming query, (Hemington, column 4).
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
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.
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
