MLOps

MLOps applies DevOps-like practices to machine learning: versioning, testing, deployment, monitoring, and lifecycle management of models.

The engineering discipline of keeping ML models reliable from notebook to production.

MLOps covers experiment tracking, feature stores, model registries, CI/CD for ML, serving, and monitoring for drift and performance.

Models fail in production without lifecycle engineering. MLOps separates notebook demos from operable systems.

• Model deployment pipelines
• Monitoring and rollback
• Feature consistency between train and serve

• Improves reliability and auditability
• Speeds safe iteration

• Tool sprawl
• Needs organizational process, not only software

• Buying an MLOps tool does not replace model validation

Is MLOps the same as AI governance?

Overlapping but distinct: MLOps is engineering lifecycle; governance includes policy, risk, and accountability.

• NIST AI RMF

• Industry MLOps maturity literature and SRE practices adapted to ML