hybrid-retrieval - Combine Vector and Keyword Search for Better RAG
Combine dense vector search and sparse keyword retrieval for improved RAG results
Tags
Updated: 2026-03-26Capabilities
Typical Inputs
Typical Outputs
What this skill does
- perform dense vector search
- execute sparse keyword search
- merge retrieval results
- rerank search results
- tune fusion weights
Inputs
- vector store
- corpus documents
- query text
- sparse encoder
- reranker model
Outputs
- merged search results
- ranked documents
- retrieval metrics
Requirements
- vector store with embeddings
- BM25 encoder
- LangChain framework
- Pinecone or Weaviate