Role-Specific LLM Responses via Nested Retrieval Augmented Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional Large Language Models (LLMs) generate similar responses to similar queries from users with different roles or ranks within an enterprise, leading to irrelevant results due to lack of context and role-specific information.
Innovation Solution
A method and system that utilize a nested Retrieval Augmented Generation (n-RAG) technique to generate user role-specific responses by combining user queries with contextual information from multiple knowledge repositories, organized based on predefined user roles, and converting this information into vector embeddings for processing by LLMs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional LLMs generate responses to user queries, then responses can be provided to all users, but the responses are generic and irrelevant to specific user roles
Solution Approach 1:
The patent segments the knowledge base into multiple knowledge repositories, each dedicated to a specific user role. This segmentation allows the system to retrieve only role-relevant information, making responses adaptable to different user roles while managing complexity through organized, modular knowledge storage.
Solution Approach 2:
The patent introduces an intermediary retrieval system that acts as a mediator between the user query and the LLM. This intermediary retrieves contextual information from role-specific knowledge repositories and combines it with the query before passing to the LLM, enabling response customization without directly modifying the LLM itself.
2Reliability
If role-specific contextual information is added to user queries, then relevant and tailored responses can be generated, but the processing complexity and time increase
Solution Approach 1:
The patent performs preliminary action by pre-organizing knowledge into role-specific repositories and pre-retrieving contextual information based on user roles before the LLM generates responses. This preparation work is done in advance, allowing the LLM to focus on generating responses rather than searching for relevant information, thus reducing overall response generation time while maintaining high relevance.
3Adaptability or versatility
If all user queries are processed with the same level of detail, then consistency is maintained, but senior executives receive excessive information while junior employees receive insufficient information
Solution Approach 1:
The patent applies local quality by providing different levels of information detail to different user roles based on their specific needs. Senior executives receive high-level summaries and strategic insights from their role-specific knowledge repository, while junior employees receive detailed operational guidance. This localized customization of information quality enhances user experience without compromising system consistency.
Data Source
AI summary
This disclosure relates to method and system for generating user role-specific responses through Large Language Models (LLMs). The method includes receiving a user query from a user account. The user account is associated with a user role from a plurality of user roles. The method includes combining, based on the user role, the user query with contextual information corresponding to the user query to obtain a combined query. The method includes inputting the combined query to an LLM. The method includes generating a user role-specific response corresponding to the combined query through the LLM. The method includes rendering the user role-specific response to a display of a user device associated with the user account.