ML MLOps

Overview

Use this skill for operational machine learning: experiment tracking, reproducibility, orchestration, registries, CI/CD, model deployment governance, monitoring, drift response, and retraining. MLOps turns notebooks and scripts into auditable, repeatable systems with clear ownership and rollback.

MLOps Invariants

Every production ML workflow should answer:

• Which data, code, config, environment, and hardware produced this model?
• Which metrics, slices, and tests justified promotion?
• Where is the model artifact stored, and how can it be rolled back?
• What validates incoming data and serving features?
• What monitors quality, drift, latency, cost, safety, and fairness?
• Who owns incidents, retraining, approvals, and deprecation?

If any answer is missing, fix the lifecycle before adding more infrastructure.

Experiment Tracking