Essential session storage is required to sign in. Optional analytics remain off unless you allow them.
Deploy a model as a service with automation and observability. Containerize a prediction API (FastAPI/BentoML), deploy it, and put it behind a CI/CD pipeline that runs DATA tests and MODEL quality gates before promoting a model from the registry (block promotion if it doesn't beat a baseline). Add MONITORING that tracks operational metrics (latency, errors) AND detects input DATA DRIFT versus a training reference (e.g. with Evidently), with alerting. Demonstrate a safe deployment (canary or shadow) and a rollback. Document the trigger for retraining.
Containerized prediction service; CI/CD with data + model quality gates that block a bad model from promotion; registry as source of truth.
Operational metrics AND data-drift detection vs a training reference, with alerting; understands models decay silently.
A canary/shadow rollout and a demonstrated rollback; offline metrics not blindly trusted.
A defined retraining trigger (schedule/drift/performance) and how it re-runs the gates.
Use this README structure to make the work reviewable.
DevPath fetches a read-only, size-capped snapshot. Submitted code is never executed.
# A served model with CI/CD and monitoring ## Goal Deploy a model as a service with automation and observability. Containerize a prediction API (FastAPI/BentoML), deploy it, and put it behind a CI/CD pipeline that runs DATA tests and MODEL quality gates before promoting a model from the registry (block promotion if it doesn't beat a baseline). Add MONITORING that tracks operational metrics (latency, errors) AND detects input DATA DRIFT versus a training reference (e.g. with Evidently), with alerting. Demonstrate a safe deployment (canary or shadow) and a rollback. Document the trigger for retraining. ## Acceptance criteria - [ ] Serving & CI/CD: Containerized prediction service; CI/CD with data + model quality gates that block a bad model from promotion; registry as source of truth. - [ ] Monitoring & drift: Operational metrics AND data-drift detection vs a training reference, with alerting; understands models decay silently. - [ ] Safe deployment: A canary/shadow rollout and a demonstrated rollback; offline metrics not blindly trusted. - [ ] Retraining plan: A defined retraining trigger (schedule/drift/performance) and how it re-runs the gates. ## Verification - [ ] Document installation and run commands. - [ ] Record automated checks and their exact commands. - [ ] Add representative output, screenshots, or a short demo where useful. - [ ] Confirm failure paths and known constraints. ## Decision log ### Decision title - Context: - Choice: - Alternatives considered: - Tradeoffs: ## Evidence - Link each rubric criterion to the file, test, or artifact that demonstrates it. ## Known limitations - List what is intentionally out of scope and what should be improved next.