New: Roadmaps ordered paths through our cheat sheets and flashcards, so you always know what to study next.
Explore themSee what's new on GitHubFrom training your first model to serving, monitoring, and defending it in production.
A 12-step learning path. Follow it in order, or jump to what you need.
For developers and data practitioners who want to move from experimenting with models to shipping and running them in production. Plan on about 7 to 12 weeks at a few hours a week, moving quickly through core ML theory into training, evaluation, deployment, and monitoring. This path stays light on classic-algorithm depth and math foundations, since Machine Learning Foundations already covers that ground step by step, and it stops short of the deep pipeline, registry, and feature-store specialization that MLOps Engineer builds on top. By the end you can train and evaluate a model with scikit-learn, package and serve it as a real API, and monitor it in production for drift before it quietly stops working.
Expected: working Python and the basics of statistics and linear algebra. Helpful but not required: SQL and a cloud platform account (AWS, GCP, or Azure).
Read this in one sitting and you'll recognize which learning paradigm any dataset calls for, the exact judgment feature engineering and model evaluation build on next.
Turns the raw columns step 1 talked about into numbers a model can actually learn from, and hands scikit-learn's pipelines the encoding judgment they lean on later.
This is where the bias-variance tension from step 1 stops being abstract and starts deciding whether you trust a score; expect to keep circling back to these metrics once real training starts next section.
You can train a first model and read its evaluation metrics without being fooled by a single flattering accuracy number. Next up: real training power with scikit-learn and deep learning.
Finish this section to unlock.
+100 XP
Puts the algorithms step 1 named and the metrics step 3 taught you to read into one Python API, so you can train and compare models in an afternoon instead of coding each by hand.
Goes under the hood of what scikit-learn's fit() hides in step 4: gradient descent, learning-rate schedules, and the regularization calls that decide whether a model converges or stalls.
Extends the optimization mechanics from step 5 to multi-layer networks, the architecture family behind most of the production vision and language models you'll deploy in section 3.
Take this if you're chasing roles where the last few points of accuracy matter: search strategies that go past the manual tweaking scikit-learn's defaults leave on the table.
You can pick an algorithm, tune its training with the right optimizer and schedule, and reach for deep learning when the data calls for it - a few minutes of due flashcards on steps 1-3 keeps that vocabulary sharp while you build on it here. Next up: getting one of these models in front of real users.
Finish this section to unlock.
+100 XP
Packages the models you've spent sections 1 and 2 training into an actual service: containers, serving frameworks, and the API layer that turns a notebook into something other systems can call.
Reframes the deployment from step 8 as an ongoing discipline instead of a one-off export - this is where 'the model works' stops meaning 'the job is done,' and teams that skip it get burned by silent decay.
Answers the question step 9 raises but doesn't resolve: how do you actually know the model has quietly stopped matching the world it was trained on.
Aim here if you're heading toward a platform team feeding features to dozens of models at once: the training-serving consistency problem that causes the sneakiest production bugs.
Pulls every earlier step into one diagram: the features from step 2, the serving from step 8, and the monitoring from step 10, so you can own the whole pipeline in a system-design interview or a real architecture review.
You can take a trained model from notebook to served endpoint, keep it healthy with monitoring and MLOps discipline, and defend the whole pipeline in a system-design conversation. That's the Machine Learning Engineer badge, earned.
Finish this section to unlock.
+100 XP
Finish every required step, at least 70% of them genuinely done (not skipped), to earn this badge and 500 XP.