MLOps: Continuous delivery and automation pipelines in machine learning

Abstraction: Google Cloud reference guide for three MLOps maturity levels with CI/CD/CT pipelines

Key points:

• MLOps extends DevOps with Continuous Training (CT) — unique to ML — automating model retraining when data distributions shift (concept drift)
• Three maturity levels: Level 0 (fully manual), Level 1 (automated ML pipeline with CT), Level 2 (full CI/CD pipeline automation)
• Only a small fraction of an ML system is ML code; surrounding infrastructure (data validation, serving, monitoring) dominates
• Feature store is a key Level 1 component: centralizes feature definitions, prevents training-serving skew, and supports both batch and low-latency serving
• Metadata management tracks pipeline versions, execution times, artifact pointers, and evaluation metrics for reproducibility and rollback
• ML pipeline triggers include: on demand, scheduled, new data arrival, model performance degradation, and significant data distribution changes

Connections: Google Cloud · Mlops · Continuous Delivery · ML Pipelines · Feature Store

Source: https://cloud.google.com/architecture/mlops-continuous-delivery-and-automation-pipelines-in-machine-learning