In this workshop, I will prepare trustworthy data in Delta, implement vector and keyword retrieval with confidence thresholds, apply grounded-answer patterns (cite-or-fail, source allow/deny) and add input/output filters for safety and PII.

You’ll wire up eval gates that run in CI/CD, capture telemetry (prompts, sources, decisions), and push actionable metrics to a simple ops dashboard. I will also cover least-privilege tool use and per-identity permissions to prevent “over-powered” assistants.

You’ll leave with a reference repo, notebooks, and checklists to deploy assistants that are accurate, auditable, and cost-aware.