Reranking
Summary
I kept seeing reranking in RAG discussions and finally clicked that it names a specific technique. Your fast initial retrieval grabs a batch of candidate documents. Then a slower, sharper model re-scores just those and reorders them before the LLM ever sees them. That second pass is where you get to weigh recency or source authority, the kind of scoring that would melt your servers if you ran it across the whole corpus.
On Reranking
I kept seeing the term reranking in RAG retrieval discussions, and it finally clicked that it names a specific job rather than a loose idea. After your initial retrieval grabs documents, a second pass re-scores and reshuffles them with something smarter than whatever yanked them in the first place.
The mechanics are two-stage. First you grab the top-K results with something fast, usually embedding similarity, which measures how close documents sit in meaning space. Then you run a beefier scoring model over just those candidates, since it's more accurate but far too expensive to run against everything. That model reorders the shortlist, and the reordered set is what goes to the LLM to generate the final response.
The point is that each stage plays a different game. Your first pass leans on semantic similarity because it has to be cheap. The rerank pass, working on a handful of candidates instead of the whole corpus, can afford to weigh recency, source authority, or task-specific relevance that would melt your servers at full scale.
In more depth, this one is pretty good: Rerankers - Pinecone