Query response generation by integrating adaptive retrieval-augmented generation with large language model

A system for generating a query response for a query using dynamic adaptive retrieval-augmented generation (RAG) pipeline comprising:

• at least one memory storing instructions;
• and
• at least one processor communicatively coupled with the at least one memory and configured to execute the stored instructions, wherein the instructions, when executed by the at least one processor, cause the at least one processor to perform operations comprising:
• receiving the query for generating the query response using a large language model (LLM);
• determining a type of question to be answered for the received query;
• converting, the query into a machine-readable format that is interpreted and executed by a database;
• translating the query based upon the type of question to be answered,
• wherein translating the query comprises determining an index level from different index levels corresponding to the type of question to be answered for the query, wherein the index level is one of:
• a sentence level index, a paragraph level index, a page level index, and/or a document level index;
• routing the query to the database to identify a preconfigured number of relevant results, wherein the preconfigured number of relevant results and the database are selected based upon the type of question to be answered;
• generating a first completion response for the query in a first iteration, by processing the query and the preconfigured number of relevant results retrieved from the database using the LLM;
• evaluating, based upon a predetermined set of parameters, the first completion response to identify quality of the first completion response;
• determining, the quality of the first completion response fails to satisfy criteria indicated by the predetermined set of parameters or a quality score assigned for the first completion response is less than or equal to a predetermined quality threshold;
• iteratively modifying, the RAG pipeline by determining an action change and/or a configuration change based on the quality of the first completion response,
• wherein the RAG pipeline is modified by:
• reconfiguring a search indexing module to dynamically select the index level from the different index levels for retrieving the relevant results for the query, independent of a relationship between the different index levels, and
• selecting, a subset of modules from a plurality of modules of the RAG pipeline, and disabling, unselected modules among the plurality of modules to generate a second completion response, based on the quality of the first completion response;
• and
• generating, the second completion response using the modified RAG pipeline for the query and the preconfigured number of relevant results retrieved from the database in a second iteration using the LLM, wherein the second completion response satisfies criteria indicated by the predetermined set of parameters.

A non-transitory computer-readable media (CRM) having instructions stored thereon, which when executed by at least one processor of at least one computing device, to generate a query response for a query using dynamic adaptive retrieval-augmented generation (RAG) pipeline by performing operations comprising:

• receiving a query for generating the query response using the large language model (LLM);
• determining a type of question to be answered for the received query;
• converting, the query into a machine-readable format that is interpreted and executed by a database;
• translating the query based upon the type of question to be answered,
• wherein translating the query comprises determining an index level from different index levels corresponding to the type of question to be answered for the query, wherein the index level is one of:
• a sentence level index, a paragraph level index, a page level index, and/or a document level index;
• routing the query to the database to identify a preconfigured number of relevant results, wherein the preconfigured number of relevant results and the database are selected based upon the type of question to be answered;
• generating a first completion response for the query in a first iteration, by processing the query and the preconfigured number of relevant results retrieved from the database using the LLM;
• evaluating, based upon a predetermined set of parameters, the first completion response to identify quality of the first completion response;
• determining, the quality of the first completion response fails to satisfy criteria indicated by the predetermined set of parameters or a quality score assigned for the first completion response is less than or equal to a predetermined quality threshold;
• iteratively modifying, the RAG pipeline by determining an action change and/or a configuration change based on the quality of the first completion response,
• wherein the RAG pipeline is modified by:
• reconfiguring a search indexing module to dynamically select the index level from the different index levels for retrieving the relevant results for the query, independent of a relationship between the different index levels, and
• selecting, a subset of modules from a plurality of modules of the RAG pipeline, and disabling, unselected modules among the plurality of modules to generate a second completion response, based on the quality of the first completion response;
• and
• generating, the second completion response using the modified RAG pipeline for the query and the preconfigured number of relevant results retrieved from the database in a second iteration using the LLM, wherein the second completion response satisfies criteria indicated by the predetermined set of parameters.