Reranking is a second retrieval stage in which a dedicated model deeply scores each candidate document against the query and reorders them by true relevance: ca
A second retrieval stage in which a dedicated model deeply scores each candidate document against the query and reorders them by true relevance, catching what fast first-stage search (vector or keyword) ranked poorly.
Usually yes. Adding a reranker is frequently the single largest quality upgrade available to a RAG system: first-stage retrieval optimizes for speed over precision, and the reranker buys the precision back at the cost of a little latency.