Detailed Action
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
This Office Action is in response to claims filed on 07/10/2025.
Claims 1-18 are pending; claims 1 and 10 are independent.
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 of this title, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-4, 6, 8-13 and 15 are rejected under 35 U.S.C. 103(a) as being unpatentable over Sörstrand et al., “Generative AI Applied to Technology Strategy-Using Open-source Large Language Models In an Agentic Way” (Sörstrand), in view of Huang, Pub. No.: US 2026/0127192 A1 (Huang).
Claim 1. Sörstrand discloses:
A retrieval augmented generation (RAG) computing module for generating an enhanced response to a user query, the module comprising: a processing unit; and a non-transitory media readable by the processing unit, the media storing instructions that when executed by the processing unit causes the processing unit to:
retrieve a plurality of documents from a database; (large documents are retrieved for chunking: sec. 2.3.2, “Unlike the single-level document splitting described in Section 2.3.1, the Parent Document Retriever introduces a two-tiered splitting mechanism, as seen in Figure 2.3. Initially, large documents are divided into what are termed “parent documents”. For instance, a multi-paged document might be divided into individual pages, each considered a parent document. These parent documents are subsequently subdivided into smaller segments referred to as “child chunks”)
identify and extract relevant context from each of the retrieved documents and append the relevant context to each respective retrieved document; (sec. 2.3.2, a link from a child to its parent is a domain specific query because it is used for reaching the parent for preserving the context of the information: “Importantly, each child chunk retains a reference back to its parent document, preserving the context of the information”)
apply a two-layer chunking process to each of the retrieved documents to generate a plurality of document chunks, wherein each of the document chunks include the appended domain-specific query and a relevant context associated with the retrieved document from which the document chunks were generated; (sec. 2.3.2, “Unlike the single-level document splitting described in Section 2.3.1, the Parent Document Retriever introduces a two-tiered splitting mechanism, as seen in Figure 2.3. Initially, large documents are divided into what are termed “parent documents”. For instance, a multi-paged document might be divided into individual pages, each considered a parent document. These parent documents are subsequently subdivided into smaller segments referred to as “child chunks”)
convert each of the document chunks into vector embeddings using an embedding model, and store the vector embeddings in a vector database; (“Each child chunk is embedded using a text embedding model that encodes the semantic meanings into the vectors”)
retrieve vector embeddings from the vector database that have similarity scores above a predetermined score, wherein a similarity score of each vector embedding is determined based on a measure of similarity between the vector embedding and a vector representation of the user query; (sec. 2.3.2, “During the retrieval phase, the process diverges from typical RAG methods by conducting the similarity search at the level of the child chunks rather than the parent documents. However, instead of returning the child chunks directly, the method retrieves the entire parent document of the highest-ranking child chunks. This strategy enhances the search’s precision by focusing on the most relevant and detailed part of text while still delivering comprehensive information by returning the full parent document”)
recursively retrieve full documents associated with the retrieved vector embeddings; and (sec. 2.3.2, “During the retrieval phase, the process diverges from typical RAG methods by conducting the similarity search at the level of the child chunks rather than the parent documents. However, instead of returning the child chunks directly, the method retrieves the entire parent document of the highest-ranking child chunks. This strategy enhances the search’s precision by focusing on the most relevant and detailed part of text while still delivering comprehensive information by returning the full parent document”)
generate, using a first large language model (LLM), the enhanced response to the user query based on the retrieved full documents and the user query. (Response to the user query is generated: sec. 2.3, “Using the retrieved information, the LLM then generates a response tailored to the user’s query. This response is informed by the most current and contextually relevant data, significantly enhancing the model’s utility and accuracy”; sec. 2.3.3, “The advantage of the Parent Document Retriever lies in its ability to refine the granularity of the search process. By focusing on smaller segments of text for indexing and retrieval, this method ensures that the semantic search is highly precise. Meanwhile, by returning the parent documents, it maintains the breadth necessary for the generation phase, ensuring that the responses are informative and contextually rich without compromising on the depth or relevance of the information provided”; sec. 3.3.2, “With the RAG and database developed, it was now possible to upload a document to the database and retrieve pages of information similar to the search query”)
Sörstrand did not specifically disclose but Huang discloses generate a domain specific query for each of the retrieved documents and append the generated domain-specific query to each of the respective retrieved documents; (¶¶ 24-26, “the system can use an LLM trained to identify a plurality of segments of data based on the uniqueness criterion (e.g., domain-specific). That is, the LLM can be trained to extract segments of the documents that correspond to domain-specific applications or documents and satisfy predefined criteria. For example, these portions can include technical terms or private vocabulary not commonly found in general language datasets. The system can also generate a plurality of queries corresponding to the extracted portions, where at least one portion of the plurality of portions can satisfy the uniqueness criterion. For example, the LLM can be used to generate the plurality of queries based on the content and context of the extracted plurality of portions (e.g., relationship of terms within the extracted data)”)
Sörstrand sec. 2.3.2, discloses that “each child chunk retains a reference back to its parent document, preserving the context of the information” meaning each parent can be reached/queried by the provided link. It would have been obvious before the effective filling date of the claimed invention to a person having ordinary skill in the art to combine the applied references for disclosing generate a domain specific query for each of the retrieved documents and append the generated domain-specific query to each of the respective retrieved documents because doing so would provide for an alternative for reaching a parent through a child.
Claim 10. Sörstrand discloses:
A method for generating an enhanced response to a user query using a retrieval augmented generation (RAG) computing module, the method comprising:
retrieving a plurality of documents from a database; (large documents are retrieved for chunking: sec. 2.3.2, “Unlike the single-level document splitting described in Section 2.3.1, the Parent Document Retriever introduces a two-tiered splitting mechanism, as seen in Figure 2.3. Initially, large documents are divided into what are termed “parent documents”. For instance, a multi-paged document might be divided into individual pages, each considered a parent document. These parent documents are subsequently subdivided into smaller segments referred to as “child chunks”)
identifying and extract relevant context from each of the retrieved documents and append the relevant context to each respective retrieved document; (sec. 2.3.2, a link from a child to its parent is a domain specific query because it is used for reaching the parent for preserving the context of the information: “Importantly, each child chunk retains a reference back to its parent document, preserving the context of the information”)
applying a two-layer chunking process to each of the retrieved documents to generate a plurality of document chunks, wherein each of the document chunks include the appended domain-specific query and a relevant context associated with the retrieved document from which the document chunks were generated; (sec. 2.3.2, “Unlike the single-level document splitting described in Section 2.3.1, the Parent Document Retriever introduces a two-tiered splitting mechanism, as seen in Figure 2.3. Initially, large documents are divided into what are termed “parent documents”. For instance, a multi-paged document might be divided into individual pages, each considered a parent document. These parent documents are subsequently subdivided into smaller segments referred to as “child chunks”)
identifying and extracting relevant context from each of the document chunks and appending the relevant context to each respective document chunk; (sec. 2.3.2, sematic meaning of each chunk is encoded into a vector: “Unlike the single-level document splitting described in Section 2.3.1, the Parent Document Retriever introduces a two-tiered splitting mechanism, as seen in Figure 2.3. Initially, large documents are divided into what are termed “parent documents”. For instance, a multi-paged document might be divided into individual pages, each considered a parent document. These parent documents are subsequently subdivided into smaller segments referred to as “child chunks”. Each child chunk is embedded using a text embedding model that encodes the semantic meanings into the vectors. This two-level hierarchical structure ensures that the indexing captures a more granular level of detail, potentially increasing the accuracy and relevance of the retrieved information”)
converting each of the document chunks into vector embeddings using an embedding model, and store the vector embeddings in a vector database; (“Each child chunk is embedded using a text embedding model that encodes the semantic meanings into the vectors”)
retrieving vector embeddings from the vector database that have similarity scores above a predetermined score, wherein a similarity score of each vector embedding is determined based on a measure of similarity between the vector embedding and a vector representation of the user query; (sec. 2.3.2, “During the retrieval phase, the process diverges from typical RAG methods by conducting the similarity search at the level of the child chunks rather than the parent documents. However, instead of returning the child chunks directly, the method retrieves the entire parent document of the highest-ranking child chunks. This strategy enhances the search’s precision by focusing on the most relevant and detailed part of text while still delivering comprehensive information by returning the full parent document”)
recursively retrieving full documents associated with the retrieved vector embeddings; and (sec. 2.3.2, “During the retrieval phase, the process diverges from typical RAG methods by conducting the similarity search at the level of the child chunks rather than the parent documents. However, instead of returning the child chunks directly, the method retrieves the entire parent document of the highest-ranking child chunks. This strategy enhances the search’s precision by focusing on the most relevant and detailed part of text while still delivering comprehensive information by returning the full parent document”)
generating, using a first large language model (LLM), the enhanced response to the user query based on the retrieved full documents and the user query. (Response to the user query is generated: sec. 2.3, “Using the retrieved information, the LLM then generates a response tailored to the user’s query. This response is informed by the most current and contextually relevant data, significantly enhancing the model’s utility and accuracy”; sec. 2.3.3, “The advantage of the Parent Document Retriever lies in its ability to refine the granularity of the search process. By focusing on smaller segments of text for indexing and retrieval, this method ensures that the semantic search is highly precise. Meanwhile, by returning the parent documents, it maintains the breadth necessary for the generation phase, ensuring that the responses are informative and contextually rich without compromising on the depth or relevance of the information provided”; sec. 3.3.2, “With the RAG and database developed, it was now possible to upload a document to the database and retrieve pages of information similar to the search query”)
Sörstrand did not specifically disclose but Huang discloses generating a domain specific query for each of the retrieved documents and appending the generated domain-specific query to each of the respective retrieved documents; (¶¶ 24-26, “the system can use an LLM trained to identify a plurality of segments of data based on the uniqueness criterion (e.g., domain-specific). That is, the LLM can be trained to extract segments of the documents that correspond to domain-specific applications or documents and satisfy predefined criteria. For example, these portions can include technical terms or private vocabulary not commonly found in general language datasets. The system can also generate a plurality of queries corresponding to the extracted portions, where at least one portion of the plurality of portions can satisfy the uniqueness criterion. For example, the LLM can be used to generate the plurality of queries based on the content and context of the extracted plurality of portions (e.g., relationship of terms within the extracted data)”)
Sörstrand sec. 2.3.2, discloses that “each child chunk retains a reference back to its parent document, preserving the context of the information” meaning each parent can be reached/queried by the provided link. It would have been obvious before the effective filling date of the claimed invention to a person having ordinary skill in the art to combine the applied references for disclosing generating a domain specific query for each of the retrieved documents and appending the generated domain-specific query to each of the respective retrieved documents because doing so would provide for an alternative for reaching a parent through a child.
Claim 2. The RAG computing module according to claim 1, wherein before the instructions to generate the enhanced response using the first LLM, the instructions further comprise additional instructions for directing the processing unit to:
select an optimal document, using a second LLM, from the retrieved full documents; and classify the selected document as the full document. (Sörstrand, an optimal document is a document selected through the highest-ranking child chunk: 2.3.2, “instead of returning the child chunks directly, the method retrieves the entire parent document of the highest-ranking child chunks. This strategy enhances the search’s precision by focusing on the most relevant and detailed part of text while still delivering comprehensive information by returning the full parent document”)
Claim 11 is rejected under the same rationale as above.
Claim 3. The RAG computing module according to claim 1, wherein the instructions to recursively retrieve the full documents associated with the retrieved vector embeddings comprises instructions for directing the processing unit to:
analyze the document chunks associated with the retrieved vector embeddings using a generative language model to identify a document chunk most relevant to the user query; and recursively retrieve at least a full document corresponding to the identified document chunk. (Sörstrand, a full document is retrieved based on the most relevant chunk: “During the retrieval phase, the process diverges from typical RAG methods by conducting the similarity search at the level of the child chunks rather than the parent documents. However, instead of returning the child chunks directly, the method retrieves the entire parent document of the highest-ranking child chunks. This strategy enhances the search’s precision by focusing on the most relevant and detailed part of text while still delivering comprehensive information by returning the full parent document”)
Claim 12 is rejected under the same rationale as above.
Claim 4. The RAG computing module according to claim 1, wherein before the instructions to convert each of the document chunks into the vector embeddings, the instructions further comprise additional instructions for directing the processing unit to:
generate, for each document chunk, a sub-query specific to content of the document chunk using a third LLM, wherein the sub-query is generated based on information contained in the document chunk and all data appended to the document chunk; and append the generated sub-query to the document chunk. (Sörstrand, wherein a parent/full document is reachable from a most relevant chunk; and Huang, ¶ 22, wherein each chunk is associated with a particular query: “The system can generate a training dataset from the domain-specific documents, such as by performing chunking of the documents, and causing a language model to generate queries ( e.g., questions) corresponding to chunks of documents”)
Claim 13 is rejected under the same rationale as above.
Claim 6. The RAG computing module according to claim 1, wherein the predetermined score comprises a similarity score above 0.76. (Sörstrand, wherein similarity score is configurable: 2.3.1, “The system then calculates a similarity score between the query embedding and each indexed chunk, typically using cosine similarity to assess the alignment of their meanings”)
Claim 15 is rejected under the same rationale as above.
Claim 8. The RAG computing module according to claim 1, whereby the instructions to retrieve the plurality of documents from the database further comprise instructions for directing the processing unit to:
retrieve the plurality of documents; and segment each of the plurality of documents into smaller and simpler documents using a large language model filter. (Sörstrand, sec. 2.3.2, “Unlike the single-level document splitting described in Section 2.3.1, the Parent Document Retriever introduces a two-tiered splitting mechanism, as seen in Figure 2.3. Initially, large documents are divided into what are termed “parent documents”. For instance, a multi-paged document might be divided into individual pages, each considered a parent document. These parent documents are subsequently subdivided into smaller segments referred to as “child chunks”; Huang, ¶ 43, “The tokenizer 108 can perform the filtering and/or extracting by prompting at least one neural network with a context (e.g., "Please identify the domain-specific terms related to "chips" and "GPU" in the text above and provide the output”)
Claim 17 is rejected under the same rationale as above.
Claim 9. The RAG computing module according to claim 1, wherein the identification and the extraction of the relevant context from each of the document chunks and the generation of the domain specific query for each of the retrieved documents are performed using a generative large language model. (Sörstrand, Abs., “This thesis explores the application of generative AI in technology strategy, specifically through the use of open-source Large Language Models (LLMs) to develop an agent capable of understanding documents and providing insightful analysis”, Huang, ¶ 63, “the language model(s) 112 can be trained and/or updated to generate contextually relevant questions based on extracted document segments. The language model(s) 112 can be or include a transformer-based model (e.g., a generative pre-trained transformer (GPT) model)”)
Claim 18 is rejected under the same rationale as above.
Claims 5 and 14 are rejected under 35 U.S.C. 103(a) as being unpatentable over Sörstrand and Huang as applied to claim 1 and 10 above in view of Маринич, Антон, “Application of Retrieval-Augmented Generation for Legal Documents” (Marynych).
Claim 5. Sörstrand as modified disclosed the RAG computing module according to claim 1; Sörstrand as modified did not specifically disclose but Marynych discloses, wherein the plurality of document chunks each comprise 512 tokens. (Marynych, p. 33, “In my experiment the bigger chunks have size of 1024 tokens and then there are sub-chunks with sizes of 128, 256 and 512 tokens that all point to their bigger parents”)
It would have been obvious before the effective filling date of the claimed invention to a person having ordinary skill in the art to combine the applied references for disclosing wherein the plurality of document chunks each comprise 512 tokens because doing so would provide for an explicit disclosure of a chunk size to be used as desired)
Claim 14 is rejected under the same rationale as above.
Claims 7-16 are rejected under 35 U.S.C. 103(a) as being unpatentable over Sörstrand and Huang as applied to claim 1 and 10 above in view of DeFoor et al., Patent No.: US 11,972,223 B1 (DeFoor).
Claim 7. Sörstrand as modified discloses:
The RAG computing module according to claim 1, wherein the instructions to recursively retrieve the full documents associated with the retrieved vector embeddings comprises instructions for directing the processing unit to:
analyze the document chunks associated with the retrieved vector embeddings using a generative language model to identify a document chunk most relevant to the user query; and (Sörstrand, 2.3.1, “The system then calculates a similarity score between the query embedding and each indexed chunk, typically using cosine similarity to assess the alignment of their meanings”)
Sörstrand as modified did not explicitly disclose but DeFoor discloses:
retrieve and classify a blank document as a full document when the generative language model determines that none of the document chunks are relevant to the user query, wherein the first LLM is triggered to generate the enhanced response based solely on the user query and its pre-trained knowledge upon receiving the blank document classified as a full document. (DeFoor, wherein an LLM is instructed do not answer a question when the answer is not found in a document, e.g., document is blank: the answer is not found in the document, col. 37, ll. 15-16, “If there are no relevant documents, respond with an empty array”; 39. 14-19, “Please answer the Question using only information in the Clauses provided. If the Clauses do not contain the answer to the Question, never try to guess- just state that the question cannot be answered from the information provided”)
It would have been obvious before the effective filling date of the claimed invention to a person having ordinary skill in the art to combine the applied references for disclosing retrieve and classify a blank document as a full document when the generative language model determines that none of the document chunks are relevant to the user query, wherein the first LLM is triggered to generate the enhanced response based solely on the user query and its pre-trained knowledge upon receiving the blank document classified as a full document because doing so would provide for instructing an LLM not answering a user question when a relevant document is not found.
Claim 16 is rejected under the same rationale as above.
Conclusion
The prior arts made of record in PTO-326 and not relied upon are considered pertinent to applicant's disclosure.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to MOHSEN ALMANI whose telephone number is (571)270-7722. The examiner can normally be reached on M-F, 9:00 to 5:00.
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 (Al R) at http://www. us pto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Ann J. Lo can be reached on 571-272-9767. 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 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. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If 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.
/MOHSEN ALMANI/Primary Examiner, Art Unit 2159
Prosecution Insights