Your RAG Treats a 3-Year-Old Doc the Same as Yesterday’s — Here’s How to Fix It

Adding content staleness tracking, CDC-based updates, and recency-weighted retrieval to a Databricks RAG pipeline You built a RAG system. It parses PDFs, chunks them, embeds them, retrieves relevant context. It even remembers conversations across turns (if you read my previous article on Lakebase-backed memory). But there’s a subtler problem hiding in plain sight. Your Vector Search index treats every chunk equally. A firmware guide from 2022 and a hotfix bulletin from last Tuesday have the same retrieval weight - as long as they’re semantically similar to the query.