Retrieval gets you candidates; reranking gets you the right ones. Vector search returns the top-k by similarity, but the most similar is not always the most relevant, so a reranker takes those candidates and reorders them by actual relevance to the question, sharply improving what the model sees.

Hybrid search adds the other half: combine semantic (vector) search with keyword search, so you catch both meaning-based matches and exact-term matches like names, codes, and jargon that embeddings alone can miss. Together they are how naive retrieval becomes production retrieval.