Capstone Scenario Briefing

The launch team is preparing a customer-support copilot that uses Delta tables, a governed feature store, a registered model, and a model serving endpoint.

Evidence packet

• Delta tables have owners, schema expectations, and freshness monitors.
• Unity Catalog grants are group-based and lineage is visible from curated data to endpoint.
• MLflow contains the promoted model version, training metrics, and evaluation report.
• The serving endpoint has a latency SLO, fallback response, and rollback owner.
• AI Gateway usage and sampled quality checks feed the launch dashboard.

Review questions

1. Which data contracts protect downstream model behavior?
2. Which assets are governed by Unity Catalog?
3. Which signal would detect stale retrieval context?
4. Who approves wider rollout after the first internal cohort?

Passing standard

The launch is ready when data, model, serving, and governance evidence can be reviewed by someone outside the implementation team.