Developer Builds Hybrid RAG System Combining FAISS Vector Search and BM25 Retrieval

An AI engineering practitioner has documented the development of a Hybrid Retrieval-Augmented Generation (RAG) system that merges FAISS semantic vector search with BM25 keyword search to improve document retrieval accuracy. The system splits source documents into chunks, converts them into vector embeddings for FAISS indexing, and simultaneously indexes them for BM25 term-based matching. Scores from both retrieval methods are normalized and combined through weighted ranking before the top results are passed as context to a large language model. The project also incorporates an agentic layer, exposing the retrieval pipeline as a callable tool so an AI agent can query the knowledge base before generating grounded responses. Qwen2.5-72B-Instruct was used as the underlying language model, with the system designed to flag queries when the knowledge base lacks sufficient information.