ML Pipelines & Automation

What you will be able to do

• Design idempotent, reproducible pipeline steps and wire them into an orchestrated DAG that reruns cleanly after a failure
• Choose an orchestration engine (Airflow, Kubeflow Pipelines, Dagster, or Prefect) that fits your team's execution model and infrastructure
• Stand up a feature store with offline and online paths, and use point-in-time joins to kill training-serving skew
• Instrument every training run with experiment tracking so any result can be reproduced, compared, and explained months later
• Promote models through a registry with versioning, stage transitions, lineage, and model cards you can hand to an auditor
• Build a continuous integration pipeline that tests data, code, and model quality and blocks a bad model before it merges
• Automate model delivery with progressive rollout and a tested rollback path, then hand off cleanly to monitoring

What is inside

4 modules, 8 lessons. Each module ends in a graded quiz and most carry an assignment.

1. 01

   Pipeline Orchestration

   Move from ad hoc notebooks and cron jobs to orchestrated DAG workflows whose steps are idempotent, versioned, and safe to rerun. You will learn to make each step reproducible, then compare Airflow, Kubeflow Pipelines, Dagster, and Prefect and choose an execution model that fits your team.

   2 lessons · 5 quiz questions

2. 02

   Feature Stores and Data Pipelines

   A feature is only useful if it means the same thing in training and in production. This module builds the feature store from the ground up: why the offline and online stores exist, how point-in-time joins keep training data honest, and how materialization and shared feature definitions in Feast eliminate training-serving skew. You will learn to define features as versioned code and reuse them across models instead of recomputing them in every notebook.

   2 lessons · 5 quiz questions · assignment

3. 03

   Experiment Tracking and the Model Registry

   An unlogged experiment is a result you cannot defend or repeat. This module shows how to capture parameters, metrics, and artifacts so any run is reproducible, then how to promote vetted models through a registry with versioning, stages, and lineage. The registry becomes the single source of truth for what is approved to deploy.

   2 lessons · 5 quiz questions

4. 04

   CI/CD for Machine Learning

   Software teams automated the path from commit to production years ago. This module brings that discipline to machine learning, where data and models must be treated as first-class citizens tested and shipped alongside code. You will build continuous integration that validates data and gates model quality, then continuous delivery that packages, progressively rolls out, and safely rolls back models, assembling the pieces from earlier modules into one hands-off pipeline.

   2 lessons · 5 quiz questions · assignment