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
- Incremental ingestion pipeline: trigger re-embedding only for changed/new documents (change detection via MD5 hash or document version field)
- Soft delete + re-index: mark outdated chunks as inactive; re-index updated version; run background cleanup
- Metadata timestamps: store
last_updatedwith each chunk; at query time, optionally filter to documents updated within N days - 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