# US 12,222,898 B1 · AI platform for processing and querying specified objects > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,222,898 B1 - Document ID: US12222898B1 - Status: Granted - Kind: B1 - Application number: 18/886,129 - Filed: 2024-09-16 - Granted: 2025-02-11 - Earliest claimed: 2024-09-16 - Assignees: Utech Products, Inc. - Applicants: Utech Products, Inc. - Inventors: Rakesh K. Madan; Zohair Hussain Jaffri; Jeevithan Alagurajah; Manish K Madan - Web page: https://patentgrep.com/patent/US12222898B1 ## CPC - G06F 16/148: {File search processing} - G06F 16/144: {Query formulation} - G06F 16/156: {Query results presentation} - G06F 16/185: Hierarchical storage management [HSM] systems, e.g. file migration or policies thereof (details of archiving G06F16/11) - G06F 16/9024: {Graphs; Linked lists (G06F16/9027 takes precedence)} - G06F 16/90332: {Natural language query formulation or dialogue systems} - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) ## Abstract An AI based system and method for processing and querying files. A method of processing files for an artificial intelligence (AI) querying service includes: hierarchically parsing the files into a set of hierarchically connected data chunks; generating metadata for each of the hierarchically connected data chunks, wherein the metadata includes hierarchical information; processing the hierarchically connected data chunks and metadata with an embedding model to generate vector embeddings that include the hierarchical information; generating textual summaries from the hierarchically connected data chunks; and storing the vector embeddings, textual summaries, and hierarchically connected data chunks for the AI querying service. ## Claims ### Claim 1 A system, comprising: a memory; and a processor coupled to the memory and configured to process a specified set of files for an artificial intelligence (AI) querying service according to process that includes: hierarchically parsing the specified set of files into a set of hierarchically connected data chunks; generating metadata for each of the hierarchically connected data chunks, wherein the metadata includes hierarchical information; processing the hierarchically connected data chunks and metadata with an embedding model to generate vector embeddings that include the hierarchical information; generating textual summaries from the hierarchically connected data chunks; and storing the vector embeddings, textual summaries, and hierarchically connected data chunks for the AI querying service. ### Claim 2 (depends on 1) The system of claim 1, wherein the AI querying service processes a query according to a process that includes: generating a query embedding from the query using the embedding model; retrieving a subset of hierarchically connected data chunks based on the query embedding, wherein the retrieving utilizes metadata filtering and nearest neighbor retrieval; selecting a subset of the textual summaries based on the query; submitting the query, subset of textual summaries, and subset of hierarchically connected data chunks to a large language model (LLM); and receiving a response from the LLM. ### Claim 3 (depends on 1) The system of claim 1, wherein the textual summaries are generated using a large language model. ### Claim 4 (depends on 1) The system of claim 1, further comprising an indexing system that links vector embeddings to the hierarchically connected data chunks. ### Claim 5 (depends on 2) The system of claim 2, wherein the subset of textual summaries are selected based on relevance to the query. ### Claim 6 (depends on 2) The system of claim 2, wherein the metadata filtering reduces a search space of hierarchically connected data chunks prior to implementing nearest neighbor retrieval. ### Claim 7 (depends on 6) The system of claim 6, wherein the metadata filtering limits nearest neighbor retrieval to hierarchically connected data chunks at a common hierarchical level. ### Claim 8 (depends on 2) The system of claim 2, wherein the query, subset of textual summaries, and subset of hierarchically connected data chunks are submitted to the LLM with a prompt. ### Claim 9 (depends on 1) The system of claim 1, wherein the embedding model is trained using a loss function that uses a multiple negative ranking loss. ### Claim 10 (depends on 1) The system of claim 1, wherein the specified set of files belong to a domain selected from a group consisting of: medical, legal, engineering, finance, and entertainment. ### Claim 11 A method of processing files for an artificial intelligence (AI) querying service, the method comprising: hierarchically parsing the files into a set of hierarchically connected data chunks; generating metadata for each of the hierarchically connected data chunks, wherein the metadata includes hierarchical information; processing the hierarchically connected data chunks and metadata with an embedding model to generate vector embeddings that include the hierarchical information; generating textual summaries from the hierarchically connected data chunks; and storing the vector embeddings, textual summaries, and hierarchically connected data chunks for the AI querying service. ### Claim 12 (depends on 11) The method of claim 11, wherein the AI querying service processes a query according to a process that includes: generating a query embedding from the query using the embedding model; retrieving a subset of hierarchically connected data chunks based on the query embedding, wherein the retrieving utilizes metadata filtering and nearest neighbor retrieval; selecting a subset of the textual summaries based on the query; submitting the query, subset of textual summaries, and subset of hierarchically connected data chunks to a large language model (LLM); and receiving and outputting a response from the LLM. ### Claim 13 (depends on 11) The method of claim 11, wherein the textual summaries are generated using a large language model. ### Claim 14 (depends on 11) The method of claim 11, further comprising generating indexes that link vector embeddings to the hierarchically connected data chunks. ### Claim 15 (depends on 12) The method of claim 12, wherein the subset of textual summaries are selected based on relevance to the query. ### Claim 16 (depends on 12) The method of claim 12, wherein the metadata filtering reduces a search space of hierarchically connected data chunks prior to implementing nearest neighbor retrieval. ### Claim 17 (depends on 16) The method of claim 16, wherein the metadata filtering limits nearest neighbor retrieval to hierarchically connected data chunks at a common hierarchical level. ### Claim 18 (depends on 12) The method of claim 12, wherein the query, subset of textual summaries, and subset of hierarchically connected data chunks are submitted to the LLM with a prompt. ### Claim 19 (depends on 11) The method of claim 11, wherein the embedding model is trained using a loss function that uses a multiple negative ranking loss. ### Claim 20 An artificial intelligence platform, comprising: a memory; and a processor coupled to the memory and configured to process a specified set of data objects for an AI querying service according to process that includes: hierarchically parsing the specified set of objects into a set of hierarchically connected data chunks; generating metadata for each of the hierarchically connected data chunks, wherein the metadata includes hierarchical information; processing the hierarchically connected data chunks and metadata with an embedding model to generate vector embeddings that include the hierarchical information; and storing the vector embeddings and hierarchically connected data chunks for the AI querying service; wherein the AI querying service processes a query according to a process that includes: generating a query embedding from the query using the embedding model; retrieving a subset of hierarchically connected data chunks based on the query embedding; submitting the query and subset of hierarchically connected data chunks to a large language model (LLM); and receiving a response from the LLM.