Multi-source RAG and Routing
Retrieval system for routing questions across structured, unstructured, and hybrid knowledge sources
Developing RAG systems that route user questions across multiple information sources instead of treating retrieval as a single-vector-search problem.
Current focus:
- Query routing across structured metadata, full-text search, vector search, and document-specific retrievers
- Hybrid retrieval with dense embeddings, keyword search, and cross-encoder reranking
- Structured extraction from messy source records into searchable metadata fields
- API-friendly responses that preserve citations, retrieved evidence, and optional conversational summaries
Tech: Python, LangChain, LanceDB/ChromaDB, Azure OpenAI, FastAPI, rerankers