// the find
GokuMohandas/Made-With-ML
Learn how to develop, deploy and iterate on production-grade ML applications.
A full MLOps course teaching how to build, train, evaluate, and deploy ML classification models using Ray and PyTorch. The whole pipeline—from Jupyter notebooks through CI/CD to Anyscale production deployment—is covered with a real NLP tagging task. Aimed at ML engineers and data scientists who want production habits, not just model accuracy.
The notebook-to-script progression is well-structured: you build intuition in Jupyter, then refactor into proper Python modules with tests. Covers the full MLOps stack in one repo—MLflow tracking, Ray Tune hyperparameter search, Ray Serve deployment, GitHub Actions CI/CD—without requiring you to assemble it yourself. Test suite is split into code/data/model layers, which is the right instinct and rarely done in tutorials. Pre-commit hooks and pyproject.toml config show the author actually cares about code hygiene.
Deep Anyscale dependency is a real problem—the production sections assume you're paying for Anyscale, and local alternatives are hand-waved with 'commands will be slightly different.' The example task (ML paper topic classification) is trivial enough that it obscures what actually gets hard at scale. Last meaningful push was March 2026 but the course content feels frozen around 2023-era Ray APIs; if Ray Serve's API changed, you'll hit friction. No coverage of monitoring or data drift detection despite listing it as a topic—those sections link out to other courses rather than showing code.