Hybrid Search for RAG: Combining BM25 and Vector Search for Better Retrieval

If you have ever watched a RAG pipeline confidently return the wrong chunk, there is a good chance the problem was not your prompt or your model. It was your retriever. Vector search alone has a speci