| CPC G06F 16/24542 (2019.01) [G06F 16/2237 (2019.01)] | 20 Claims |
|
1. A method of using an intelligent search agent to search a knowledge base comprising a plurality of data items indexed according to a multimodal Retrieval Augmented Generation (RAG) encoding system, said method comprising:
using an intelligent search agent comprising at least one processor and at least one generative AI system to accept search input from at least one user, interpret said search input, tokenize and vectorize at least some of said interpreted search input, thus producing at least one vectorized query for a present search;
using said intelligent search agent and said at least one vectorized query to search at least some of said knowledge base, and determine at least some data items with indexes that show significant similarity with said at least one vectorized query, thereby producing retrieved data items;
wherein said knowledge base comprises a plurality of different RAG databases, each different RAG database comprising a plurality of data items, each data item indexed by a vector generated by at least one of said at least one Generative AI systems acting upon a tokenized version of that respective data item (vector indexed data item);
wherein said intelligent search agent uses at least some of said retrieved data items, and said search input, to return at least some information from said knowledge base as response information;
wherein, for at least some of said at least one users, said intelligent search agent further uses any of said search inputs, interpreted search input, tokenized search input, said at least one vectorized queries, and said response information to create a learning database for use in any of said present search and for use in subsequent searches;
said intelligent search agent is configured to use said learning database to determine which of said plurality of different RAG databases produce retrieved data items with any of a lower or higher significant similarity over a plurality of said searches, and to produce at least one reformatted RAG database depending on said determination;
wherein said intelligent search agent is further configured to autonomously interpret search inputs, query external data sources comprising raw data obtained via web scraping and pre-processed vectorized data, and execute automated interventions based on search results as an active system component; and
wherein said intelligent search agent is further configured to apply one or more optimization algorithms to dynamically adjust search parameters based on computational efficiency constraints and to apply machine learning techniques to modify search strategies in response to prior search performance and user feedback.
|