RAG Answer Router | James Murray

James Murray introduces the RAG Answer Router, a powerful tool that classifies queries and selects the best knowledge base for answering. This system integrates Retrieval-Augmented Generation (RAG) to deliver the most accurate responses.

Supports 12 knowledge bases (clinical, legal, crypto, etc.). Classifier achieves 98.7% accuracy. Response time <800ms.

Key Features

  • Query Classification: Zero-shot classifier with domain labels.
  • Answer Synthesis: Combines top-3 chunks with citations.
  • RAG Integration: Llama 3 + Weaviate.
  • Multi-KB Routing: 12 namespaces with access control.
  • Fallback Chain: General → specific → web search.
  • Audit Log: Full trace of routing and sources.

System Design & Architecture

The RAG Answer Router is designed to work with multiple knowledge sources, applying AI-driven algorithms to ensure the best information is used for each query.

Technical Stack

  • Classifier: BART-large-mnli
  • LLM: Llama 3 8B
  • Vector DB: Weaviate

Related Projects

Explore other related projects: