Asked about Where embedding search fails

What is hybrid search?

An answer that works

Running keyword and vector search together and fusing the results, usually by rank rather than raw score since the two scoring scales are incomparable. Keyword catches exact terms, vector catches paraphrase.

Reading an answer is not being able to give one. The section that teaches this is Where embedding search fails, 3 hours of Embeddings.

Learn this properly

Also asked about Embeddings

What is an embedding?Why cosine similarity rather than Euclidean distance?Can you compare embeddings from two different models?Why does embedding search sometimes return irrelevant results?What does a reranker do?Your RAG system returns wrong answers. How do you debug it?