Made With ML

What it covers

The main MLOps course is organised into eight sections:

1. Design: setting up, product thinking and system design for an ML application.
2. Data: preparation, exploration, preprocessing and distributed data processing.
3. Model: training, experiment tracking, hyperparameter tuning, evaluation and serving.
4. Developing: moving from notebooks to scripts and a command-line interface.
5. Utilities: logging, documentation, code style and pre-commit hooks.
6. Testing: tests for code, data and models.
7. Reproducibility: versioning code, data and models.
8. Production: jobs and services, CI/CD, monitoring and data engineering.

A separate Foundations track covers notebooks, Python, NumPy, pandas and PyTorch, then ML and deep learning concepts from linear regression through neural networks, CNNs, embeddings, RNNs and transformers.

Why it stands out

Most ML courses end when the model works on the test set. This one starts there and treats the model as one part of a software system. The same example project runs through every lesson, so by the end you have a single, complete application with tests, versioning and deployment, not a collection of disconnected notebooks.

It is popular: the GitHub repository had about 49,600 stars and 7,800 forks as of September 2026, and the site says more than 40,000 developers have taken part. The course is MIT-licensed; the repository's last update was in March 2026.

Before you start

The course asks for Python fundamentals and machine learning basics, and recommends prior experience with ML in industry or academia. If you have never trained a model, do the Foundations track or a course such as the Machine Learning Specialization first.

The production lessons use Ray and Anyscale for distributed training and serving, which the author now works on. The principles transfer to other stacks, but expect to adapt some tooling if your company uses something else.

How to take it

• Clone the repository and run every lesson's code, rather than only reading.
• Rebuild the project with your own dataset after the first pass. That is where understanding really forms.
• Focus on the testing, reproducibility and CI/CD sections; these are the skills least taught elsewhere and most valued in ML engineering roles.
• Expect several weeks of steady work, more if you do the Foundations track first.

How it compares

• MLOps Zoomcamp: also free, organised as a nine-week course with homework and a community, and uses a different toolset (MLflow, Prefect, AWS). Good if you want a structured schedule.
• Designing Machine Learning Systems: the book explains the design decisions and trade-offs; Made With ML makes you implement them.
• Full Stack Deep Learning: a similar focus on production deep learning, as lectures.

Common questions

Is it free? Yes. The course content is free online; optional paid live cohorts have been offered with lessons, Q&A and compute.

Do I need a GPU? Not for most lessons. Distributed training sections can run on small clusters or be followed conceptually.

Is it still current? The principles are. Check library versions when you run the code, since the repository was last updated in March 2026.

Best for: engineers who already know how to build models and need to learn how to ship, test and maintain them in production.