// the find
GokuMohandas/mlops-course
Learn how to design, develop, deploy and iterate on production-grade ML applications.
A course repo teaching MLOps end-to-end: data engineering, training, experiment tracking with MLflow, Ray-based distributed execution, and deployment via Anyscale. It targets ML practitioners who know how to train a model but haven't wired one into a production system. The companion website (madewithml.com) is where the actual lessons live; this repo is the code artifact.
The project is structured as real software, not just notebooks — it refactors the notebook code into proper Python modules with tests, pre-commit hooks, and CI/CD pipelines. The three-layer test suite (code, data, model) is a pattern most ML projects skip and genuinely useful to copy. Ray integration for distributed training, tuning, and serving is shown with realistic CLI arguments rather than toy examples. GitHub Actions workflows for train-on-PR and deploy-on-merge are production-grade and directly usable.
The last push was August 2024 and the Anyscale dependency is load-bearing — most of the production path (jobs, services, cluster environments) assumes you're paying for Anyscale, with only a footnote pointing to self-managed Ray clusters. The dataset is a tiny 4-class text classifier (ML project tags), which means the model itself is trivial and the MLOps scaffolding can obscure whether any of this would hold up with a real workload. There's no monitoring or drift detection in the actual code — the README mentions it as a link to more madewithml content, which means the continual learning section is incomplete without the paid course.