loading
loading
A second pass that takes your top search results and reorders them by true relevance using a heavier, more accurate model.
Reranking sits after retrieval: your vector search returns the top 50 candidates fast but roughly, then a cross-encoder reranker scores each one against the query directly and pushes the genuinely relevant ones to the top. It's slower per item than embedding search, which is why you only run it on the shortlist, not the whole corpus. In RAG this is the cheapest big win — better ordering means the right context lands in the model's window instead of getting truncated away.
Plainly
Think of Reranking as a simple recipe for doing the work better. A second pass that takes your top search results and reorders them by true relevance using a heavier, more accurate model.
In practice
Use it when you need a repeatable method instead of guessing from vibes. In practice, define the owner, input, output, and failure mode before you rely on it.