Skip to content

Production Track

This track focuses on operational ML: model serving, drift detection, release decisions, and system-level tradeoffs.

  1. projects/mlops-drift-production-showcase
  2. projects/batch-vs-stream-ml-systems-showcase
  3. projects/model-release-rollout-showcase
  4. projects/demand-api-observability-showcase

Core Skills Covered

  • Monitoring feature and prediction drift.
  • Serving models with contract-aware API endpoints.
  • Canary release decisions and rollback criteria.
  • Batch vs stream pipeline tradeoffs.
  • Observability with structured logs, metrics, and traces.

Evidence Artifacts To Inspect

  • artifacts/drift/ outputs in MLOps showcase
  • openapi.json in API showcases
  • artifacts/registry/model_versions.json
  • rollout decision logs and simulation outputs

Suggested Reflection Prompts

  • Which KPI would trigger a rollback first and why?
  • What drift signal is most actionable for retraining decisions?
  • Where should alert thresholds differ between batch and online systems?