Skip to content
Quantoros

Your RAG knowledge base is updated frequently. How do you manage freshness?

LLM & AILevel: IntermediateTroubleshooting Scenarios

Your RAG knowledge base is updated frequently. How do you manage freshness?

Hints
  • Break the query into candidate retrieval, context selection and generation.
  • How would you locate the stage that lost a fact without using only one end-to-end metric?

Answer

  1. Incremental ingestion pipeline: trigger re-embedding only for changed/new documents (change detection via MD5 hash or document version field)
  2. Soft delete + re-index: mark outdated chunks as inactive; re-index updated version; run background cleanup
  3. Metadata timestamps: store last_updated with each chunk; at query time, optionally filter to documents updated within N days
  4. Two-index pattern: "current" index (recent, high quality) + "archive" index; route time-sensitive queries to current index

Source and licence: JustInCache & contributors — Awesome AI Interviews · Apache-2.0

Related questions