Overview
Training a model and shipping it are different problems, and most machine learning material stops at the first one. This course picks up where a notebook ends: turning exploratory code into something that runs reliably, is reproducible on someone else’s machine, and can be deployed and monitored rather than just demoed.
It is a working repository rather than a lecture series: reproducible environments, real pipelines and real deployment code, meant to be run and adapted rather than watched.
What it covers
Reproducible environments with pyenv and Poetry, and Makefile-driven automation so the same commands work for everyone. Structuring ML code as a proper Python package rather than a folder of notebooks. Continuous integration and deployment practices. Experiment tracking and model versioning with MLflow. Pipeline scheduling and orchestration with Apache Airflow. Serving the resulting model as an API with Flama, our own production framework.
Who it’s for
Developers and data scientists who already know how to train a model and want the operational half: the gap between a notebook that works on one machine and a system that runs reliably in production.
Built with
Python 3.12, scikit-learn, Flama, Docker, Apache Airflow and MLflow, with ruff, black, isort and pyright for code quality.