Machine Learning Zoomcamp earns its reputation by teaching the half of machine learning that most courses quietly skip, which is what happens after the model works in your notebook. Alexey Grigorev runs it the way a good engineering team works, so you learn regression, classification, and model evaluation, but you also learn to wrap a model in a Flask service, containerise it with Docker, push it to the cloud, and run it serverless, and there is a deep learning stretch with TensorFlow and Keras plus a look at Kubernetes and model serving. That deployment focus is the whole point and it is what makes the course feel like a paid bootcamp that happens to cost nothing. It is project based, you submit work and review other people's submissions, and the two capstone projects are the kind of thing you can actually put in front of an employer, which matters far more than a certificate.
The annual cohort adds deadlines, a lively Slack, and the gentle pressure that gets people to the finish line, and even self-paced you get every video and notebook on GitHub. The tradeoffs are clear and fair. It assumes you are already comfortable in Python, so a true beginner will struggle, and it is deliberately light on the underlying mathematics, so if you want to understand why gradient descent converges you will need a theory course alongside it. The self-paced route also loses the accountability that carries cohort students through the harder weeks.
None of that undercuts the value. For someone who can code and wants to be a machine learning engineer who ships, this is the most generous free course I know of, and I would happily recommend it over plenty of paid alternatives.