Switch language한국어
Back to the list

The laptop return that broke a RAG pipeline

TL;DR AI

Key summary

2 min read
  1. A user asked whether they could return a laptop purchased three weeks earlier.

  2. The agent retrieved a 2023 return-policy document, quoted a 30-day window, and told the customer to ship it back.

  3. The current policy has since changed to a 14-day window for electronics (policy update since 2023), so the retrieved document was outdated.

  4. 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.

  5. The author defines hybrid search—combining vector retrieval with structured metadata—as the needed solution.

Read the original