RAG :: PIPELINE

    What Comes After Reranking in a RAG Pipeline?

    In a RAG pipeline, the results that survive reranking are passed to the language model as context for generating the answer. Strategically, teams that have already added reranking and want further gains look beyond query-time techniques to ingestion-time optimization, improving the index itself so retrieval returns better candidates before reranking runs. Reranking refines order; ingestion-time optimization improves what is available to refine.

    The pipeline step after reranking

    Mechanically, reranking is usually the last retrieval step. The top reranked documents become the context passed to the language model, which generates the final answer.

    The strategic next lever, and why it can replace the step

    Teams add reranking to improve relevance, then find its gains plateau because reranking can only reorder what retrieval already returned. The more effective move is earlier in the pipeline: clean the index at ingestion so retrieval returns the right candidates in the first place. When the index is clean from the start, the reranking step it was compensating for often becomes unnecessary. Green Vectors applies this by eliminating redundancy at ingestion, raising first-pass relevance so reranking is optional for most workloads.

    More questions

    Not always. Reranking reorders retrieved results but cannot recover missed documents. Cleaning the index at ingestion addresses the cause rather than the symptom.
    Improving the vector index at the point data enters the system, before storage, so every query benefits from a cleaner index without per-query cost.

    Ready to go deeper?

    Request access to Kitana, our Python SDK built on Green Vectors, or get in touch with the team.