How to build a context-aware AI assistant with multiple LLMs?
A MULTI-LLM ORCHESTRATION ENGINE FOR PERSONALIZED, CONTEXT-RICH ASSISTANCE
October 15, 2024
https://arxiv.org/pdf/2410.10039This paper proposes a novel architecture for personalized AI assistants that uses multiple LLMs, a temporal graph database, and a vector database.
Key points for LLM-based multi-agent systems:
- Overcomes context window limitations: Uses a temporal graph database to store and retrieve conversation history, enabling long-term context retention.
- Integrates private data: Utilizes a vector database to efficiently store and access private user data without retraining the LLMs.
- Reduces hallucinations: Employs multiple LLMs in an orchestration engine that iteratively refines responses, increasing accuracy and minimizing irrelevant information.
- Improves personalization: Combines context from the graph database with specific data from the vector database to deliver more personalized and relevant answers.