How to Build Agentic RAG with Hybrid Search

Traditional Retrieval Augmented Generation (RAG) systems using only vector similarity can struggle with specific keywords or IDs. Hybrid search improves this by combining semantic search with keyword-based algorithms like BM25 for more accurate retrieval. The approach can be further enhanced by creating an "agentic" RAG system, which empowers an LLM to act as an agent controlling the search process. This agent can rewrite queries, iteratively fetch information, and dynamically weigh the results from vector and keyword searches to achieve superior performance.