US 12,488,259 B2
System for providing software related answer based on a trained model
Joel Hart, Evergreen, CO (US); and Douglas Lee Saferite, Lake City, FL (US)
Assigned to Continuity AI, LLC, Boulder, CO (US)
Filed by Continuity AI, LLC, Boulder, CO (US)
Filed on Feb. 8, 2024, as Appl. No. 18/437,155.
Claims priority of provisional application 63/483,905, filed on Feb. 8, 2023.
Prior Publication US 2024/0265281 A1, Aug. 8, 2024
Int. Cl. G06F 40/44 (2020.01); G06F 40/58 (2020.01); G06N 5/04 (2023.01)
CPC G06N 5/04 (2013.01) [G06F 40/44 (2020.01); G06F 40/58 (2020.01)] 21 Claims
OG exemplary drawing
 
1. A computer-implemented method for providing natural language understanding of a development process, the method being executed by an electronic processor and comprising:
receiving, from a user interface, a natural language query regarding a proprietary code base of an organization; processing the natural language query through a custom enhancement model to determine an intent of the natural language query;
providing a query/response enhancement service to enhance the query by using domain specific knowledge and a large language model (LLM);
processing the enhanced query through a trained model to determine a natural language response for the natural language query, the trained model trained with generic inputs and specific inputs, the specific inputs including information from proprietary corpora of the organization;
wherein processing the natural language query through the custom enhancement model to determine the intent of the natural language query and provide an enhanced query includes processing the natural language query with the trained model and a context of the natural language query;
enhancing the query by adding metadata filters based on a context being asked or a custom corpora about the user's architecture learned during ingestion; and
providing, through the user interface, access to the natural language response.