Intelligent history management in chatbots

A non-transitory computer readable medium (CRM) comprising computer readable program code, which when executed by a computer processor, enables the computer to perform a method for processing user queries, the method comprising:

• providing a first processed user query to a large language model (LLM), wherein the first processed user query is associated with a user;
• generating, by the LLM and based on the first processed user query, a first list of domain candidates;
• assigning, by the LLM, a first confidence score to each domain candidate in the first list of domain candidates;
• identifying a first domain candidate from the first list of domain candidates using the first confidence scores;
• making a first determination that there is a previous domain, wherein the previous domain is associated with another user query previously provided by the user;
• making a second determination, based on the first determination, that the first domain candidate is different than the previous domain;
• making a third determination, based on the second determination, that the first domain candidate does not correspond to a follow-up domain;
• switching, based on the third determination, the previous domain to the first domain candidate;
• in response to the switching:
• erasing any previous conversation history associated with the another user query previously provided by the user and providing only the first processed user query to a Retrieval Augmented Generation (RAG) application;
• and
• obtaining, from the RAG application, and based on the first processed user query, relevant information and providing the relevant information and the first processed user query to the LLM;
• and
• generating, by the LLM and in response to the obtaining, output and providing the output to the user via a user interface.

An intelligent history management system, comprising:

• a processor;
• a large language model (LLM);
• memory comprising instructions, which when executed by the processor, performs a method for switching domains using history management, the method comprising:
• obtaining a user query from a user via a user interface;
• cleaning the user query to obtain a pre-processed user query;
• extracting key features from the pre-processed user query to obtain a first processed user query;
• providing the first processed user query to the LLM, wherein the first processed user query is associated with a user;
• generating, by the LLM and based on the first processed user query, a first list of domain candidates;
• assigning, by the LLM, a first confidence score to each domain candidate in the first list of domain candidates;
• identifying a first domain candidate from the first list of domain candidates using the first confidence scores;
• making a first determination that there is a previous domain, wherein the previous domain is associated with another user query previously provided by the user;
• making a second determination, based on the first determination, that the first domain candidate is different than the previous domain;
• making a third determination, based on the second determination, that the first domain candidate does not correspond to a follow-up domain;
• switching, based on the third determination, the previous domain to the first domain candidate;
• in response to the switching:
• erasing any previous conversation history associated with the another user query previously provided by the user and providing only the first processed user query to a Retrieval Augmented Generation (RAG) application;
• obtaining, from the RAG application and based on the first processed user query, relevant information and providing the relevant information and the first processed user query to the LLM;
• and
• generating, by the LLM and in response to the obtaining, output and providing the output to the user via the user interface.