Back to index
OtherSelf paced, roughly 30 to 40 hours of material, with an optional one day live cohort·Free and open source, optional paid live cohort

Made With ML (MLOps Course by Goku Mohandas)

4.6

The course I wish existed when I first tried to put a model into production. Made With ML is not really about training models, it is about everything that surrounds them, and it is one of the best free resources anywhere for the awkward gap between a working notebook and a system that runs reliably in front of real users.

What We Liked

  • Free, open source and genuinely current, with a public GitHub repo you actually work through
  • Fills the notebook to production gap that almost every ML course skips
  • Strong on the boring, important things: testing, CI/CD, experiment tracking, reproducibility and monitoring
  • Written by a practitioner who has shipped this stuff, so the advice is opinionated and pragmatic

What Could Be Better

  • Assumes you already know Python and the basics of machine learning, so it is not a first course
  • Leans on Ray and the Anyscale stack in places, which colours some of the tooling choices
  • Being code heavy and library dependent, the repo can drift as versions move on
  • Light on model theory, since the whole point is the engineering around the model

Detailed review

Most machine learning education stops at the moment the model prints a nice accuracy number, which is precisely where the hard part begins. Made With ML, built by Goku Mohandas and now sitting under Anyscale, is one of the few free courses that starts where the others finish. The framing is production from day one. You take a real problem, and rather than lingering on model architecture you spend your time on the scaffolding that decides whether a model is actually useful: structuring the codebase like software rather than a pile of cells, writing tests for data and for the model, tracking experiments, building a training and evaluation pipeline, then wrapping the whole thing in CI/CD so changes ship safely and can be rolled back.

That emphasis is the entire value here. Anyone can follow a tutorial that fits a classifier, but very few resources teach you how to make that classifier reproducible, observable and maintainable, and this one does it clearly and with obvious hard won opinions. The lessons are readable, the accompanying repository is well organised, and because it is open source you can see exactly how everything fits together instead of trusting a slide. There are honest caveats.

This is not a beginner course, and it says so. You need to be comfortable in Python and to already understand what a model is and why you would evaluate one, otherwise the software engineering emphasis will feel like learning two hard subjects at once. There is also a natural pull toward Ray and the Anyscale ecosystem for the distributed and scaling portions, which makes sense given who maintains it, but means a little of the tooling is a particular choice rather than a neutral standard. And as with any code first course that depends on a fast moving stack, you should expect the occasional version mismatch that you have to reason through yourself, which frankly is decent practice for the job anyway.

Weighed up, for someone who can build a model and now wants to ship one, this is a standout, and the price makes the recommendation easy. Learn the engineering habits as broadly transferable, take the specific stack as one sensible implementation, and you close the exact gap that keeps a lot of capable people stuck in notebooks.

[ final ]

The verdict.

If you can already build a model but have no idea how to make it survive contact with production, this is close to the ideal next step, and the fact that it is free is almost unfair. Treat the Ray specific parts as one good way of doing things rather than the only way, and you come away with skills that most self taught ML people are missing.