# US 12,572,574 B2 · Retrieval augmented generative question and answer boosting > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,572,574 B2 - Document ID: US12572574B2 - Status: Granted - Kind: B2 - Application number: 18/640,914 - Filed: 2024-04-19 - Published: 2025-10-23 - Granted: 2026-03-10 - Earliest claimed: 2024-04-19 - Assignees: Life Technologies Corporation; Thermo Electron North America LLC; PPD Development, Lp - Applicants: Life Technologies Corporation; Thermo Electron North America LLC; PPD Development, L.P. - Inventors: Bradley Jared Larson; Thomas S. Schutz; Rachel Alane Heskin; Anthony Owen Cooper; John Christian Jameson; Lawrence Michael Lobato, Jr.; Praveen Kakumanu; Soumak Mookherjee - Web page: https://patentgrep.com/patent/US12572574B2 ## CPC - G06F 16/3344: {using natural language analysis} - G06F 16/3329: Natural language query formulation - G06F 40/40: Processing or translation of natural language (natural language analysis G06F40/20; semantic analysis G06F40/30) - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) ## Abstract Systems or techniques are provided for facilitating retrieval augmented generative question and answer boosting. In various embodiments, a system can access a plain text question regarding a scientific instrument. In various aspects, the system can generate, via a large language model that references a document-graph repository, a structured or unstructured answer for the plain text question. In various instances, the document-graph repository can comprise a plurality of document-graphs that respectively correspond to a plurality of technical documents. In various cases, for a first document-graph that corresponds to a first technical document, leaf nodes of the first document-graph can represent respective text blocks written in the first technical document, and non-leaf nodes of the first document-graph can respectively represent a document title, one or more section headings, and one or more scientific instrument identifiers written in the first technical document and beneath which the respective text blocks are nested. ## Claims ### Claim 1 A system, comprising: a processor that executes computer-executable components stored in a non-transitory computer-readable memory, wherein the computer-executable components comprise: an access component that accesses a plain text question regarding a scientific instrument; and an answer component that generates, via a large language model that references a document-graph repository, a structured or unstructured answer for the plain text question. ### Claim 2 (depends on 1) The system of claim 1, wherein the document-graph repository comprises a plurality of document-graphs that respectively correspond to a plurality of technical documents, wherein, for a first document-graph of the plurality of document-graphs that corresponds to a first technical document of the plurality of technical documents, leaf nodes of the first document-graph represent respective text blocks written in the first technical document, and wherein non-leaf nodes of the first document-graph respectively represent a document title, one or more section headings, and one or more scientific instrument identifiers written in the first technical document and beneath which the respective text blocks are nested. ### Claim 3 (depends on 2) The system of claim 2, wherein, for a first leaf node that is nested beneath one or more first non-leaf nodes of the first document-graph, a concatenation of the first leaf node with the one or more first non-leaf nodes forms a context-tagged text block, and wherein the document-graph repository comprises an embedding that semantically represents the context-tagged text block. ### Claim 4 (depends on 3) The system of claim 3, wherein the computer-executable components comprise: a search component that constructs the document-graph repository via actions comprising: document-wise execution of a trained text-to-graph model on the plurality of technical documents; or document-wise execution of the large language model on the plurality of technical documents and on one or more document-exploration prompts. ### Claim 5 (depends on 1) The system of claim 1, wherein the computer-executable components comprise: a search component that identifies, via an embedding search or a keyword search, a context-tagged text block from the document-graph repository that is related to the plain text question, generates a composed context-tagged text block by iteratively performing document-graph walking and embedding-change comparison with respect to the context-tagged text block, and generates a unified prompt based on concatenating the plain text question with the composed context-tagged text block, wherein the large language model receives the unified prompt as input and produces the structured or unstructured answer as output. ### Claim 6 (depends on 1) The system of claim 1, wherein the computer-executable components comprise: a search component that identifies some of a plurality of context-tagged text blocks from the document-graph repository via an embedding search based on the plain text question and others of the plurality of context-tagged text blocks from the document-graph repository via a keyword search based on the plain text question, and wherein the search component selects, via execution of a re-ranker, one or more highest-ranking context-tagged text blocks from the plurality of context-tagged text blocks, wherein the search component generates a unified prompt by concatenating the plain text question with the one or more highest-ranking context-tagged text blocks, and wherein the large language model receives the unified prompt as input and produces the structured or unstructured answer as output. ### Claim 7 (depends on 6) The system of claim 6, wherein the search component biases the re-ranker toward context-tagged text blocks found using the keyword search, in response to a determination that the plain text question comprises a scientific instrument identifier. ### Claim 8 (depends on 6) The system of claim 6, wherein the search component causes the large language model to respond to one or more chain-of-thought prompts for respective ones of the plurality of context-tagged text blocks, thereby yielding a plurality of chain-of-thought responses that respectively correspond to the plurality of context-tagged text blocks, wherein the re-ranker assigns respective relevance scores to the plurality of chain-of-thought responses, and wherein the search component identifies the one or more highest-ranking context-tagged text blocks based on the relevance scores. ### Claim 9 (depends on 1) The system of claim 1, wherein the answer component prompts the large language model to pay extra attention to an identifier of the scientific instrument, in response to the identifier being written in the plain text question. ### Claim 10 A computer-implemented method, comprising: accessing, by a device operatively coupled to a processor, a plain text question regarding a scientific instrument; and generating, by the device and via a large language model that references a document-graph repository, a structured or unstructured answer for the plain text question. ### Claim 11 (depends on 10) The computer-implemented method of claim 10, wherein the document-graph repository comprises a plurality of document-graphs that respectively correspond to a plurality of technical documents, wherein, for a first document-graph of the plurality of document-graphs that corresponds to a first technical document of the plurality of technical documents, leaf nodes of the first document-graph represent respective text blocks written in the first technical document, and wherein non-leaf nodes of the first document-graph respectively represent a document title, one or more section headings, and one or more scientific instrument identifiers written in the first technical document and beneath which the respective text blocks are nested. ### Claim 12 (depends on 11) The computer-implemented method of claim 11, wherein, for a first leaf node that is nested beneath one or more first non-leaf nodes of the first document-graph, a concatenation of the first leaf node with the one or more first non-leaf nodes forms a context-tagged text block, and wherein the document-graph repository comprises an embedding that semantically represents the context-tagged text block. ### Claim 13 (depends on 12) The computer-implemented method of claim 12, further comprising: constructing, by the device, the document-graph repository via actions comprising: document-wise execution of a trained text-to-graph model on the plurality of technical documents; or document-wise execution of the large language model on the plurality of technical documents and on one or more document-exploration prompts. ### Claim 14 (depends on 10) The computer-implemented method of claim 10, further comprising: identifying, by the device and via an embedding search or a keyword search, a context-tagged text block from the document-graph repository that is related to the plain text question; generating, by the device, a composed context-tagged text block by performing actions comprising iterative document-graph walking and embedding-change comparison with respect to the context-tagged text block; and generating, by the device, a unified prompt based on concatenating the plain text question with the composed context-tagged text block, wherein the large language model receives the unified prompt as input and produces the structured or unstructured answer as output. ### Claim 15 (depends on 11) The computer-implemented method of claim 11, further comprising: identifying, by the device, some of a plurality of context-tagged text blocks from the document-graph repository via an embedding search based on the plain text question and others of the plurality of context-tagged text blocks from the document-graph repository via a keyword search based on the plain text question; selecting, by the device and via execution of a re-ranker, one or more highest-ranking context-tagged text blocks from the plurality of context-tagged text blocks; and generating, by the device, a unified prompt by concatenating the plain text question with the one or more highest-ranking context-tagged text blocks, wherein the large language model receives the unified prompt as input and produces the structured or unstructured answer as output. ### Claim 16 (depends on 15) The computer-implemented method of claim 15, wherein the device biases the re-ranker toward context-tagged text blocks found using the keyword search, in response to a determination that the plain text question comprises a scientific instrument identifier. ### Claim 17 (depends on 15) The computer-implemented method of claim 15, wherein the device causes the large language model to respond to one or more chain-of-thought prompts for respective ones of the plurality of context-tagged text blocks, thereby yielding a plurality of chain-of-thought responses that respectively correspond to the plurality of context-tagged text blocks, wherein the re-ranker assigns respective relevance scores to the plurality of chain-of-thought responses, and wherein the device identifies the one or more highest-ranking context-tagged text blocks based on the relevance scores. ### Claim 18 A computer program product for facilitating retrieval augmented generative question and answer boosting, the computer program product comprising a non-transitory computer-readable memory having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to: access a natural language question asking about maintenance or operation of a scientific instrument; identify, from a document-graph repository, one or more context-tagged text blocks that are relevant to the natural language question; and generate, via execution of a large language model on a unified prompt that is based on the natural language question and on the one or more context-tagged text blocks, a natural language answer. ### Claim 19 (depends on 18) The computer program product of claim 18, wherein, in response to the natural language question containing an identifier of the scientific instrument, the processor causes the unified prompt to comprise a natural language sentence instructing the large language model to pay extra attention to the identifier. ### Claim 20 (depends on 18) The computer program product of claim 18, wherein the processor identifies the one or more context-tagged text blocks via an embedding search, a keyword search, and a re-ranker, and wherein the re-ranker respectively assigns relevance scores to the one or more context-tagged text blocks based on texts synthesized by the large language model in response to chain-of-thought prompts corresponding to the one or more context-tagged text blocks.