A retrieval demo on a clean corpus looks easy. Production RAG breaks in quieter ways: wrong chunk boundaries, stale indexes, permission leaks across tenants, confident answers with weak grounding, and evaluations that never catch regressions.

This talk covers product and architecture lessons from shipping enterprise retrieval systems (RAG, vector search, LLM orchestration) from pre-revenue to enterprise customers, plus high-stakes operational software lessons. It covers failure modes after the happy path: what to measure, what to gate before release, and how to prioritize fixes when a pilot looks fine.

Attendees get a checklist: indexing and refresh, access control, evaluation harnesses, fallback behavior, and operator UX when the system should abstain. Examples stay vendor-agnostic.