The laptop return that broke a RAG pipeline

TL;DR AI
2 min readKey summary
A user asked whether they could return a laptop purchased three weeks earlier.
The agent retrieved a 2023 return-policy document, quoted a 30-day window, and told the customer to ship it back.
The current policy has since changed to a 14-day window for electronics (policy update since 2023), so the retrieved document was outdated.
The root cause was architectural: vector similarity matched the 2023 policy well (high cosine similarity) but lacks recency and scope, producing a retrieval accuracy gap; timestamps, scopes, and permissions are structured data in columns, not in vector space.
The author defines hybrid search—combining vector retrieval with structured metadata—as the needed solution.



