Day 2 - MLOps: SageMaker Pipelines, Model Registry, CI/CD

Yesterday I said drift detection must trigger retraining. If retraining is manual, operations break. MLOps means defining data processing → training → evaluation → registration → deployment in code to make them repeatable and traceable. Today we cover defining workflows with SageMaker Pipelines, managing model versions with Model Registry, and automating deployment with CI/CD.

The Problem MLOps Solves

Manual ML: run a model once in a notebook, deploy from console by hand. Unrepeatable, untraceable, non-collaborative. MLOps solves this three ways: