A step in a RAG pipeline where initially retrieved results are re-scored and reordered for relevance before being passed to the model.
Initial retrieval often returns reasonable but imperfectly ordered results — reranking improves precision by applying a more sophisticated relevance model to just the top candidates.
