// the find
featurestoreorg/serverless-ml-course
Serverless Machine Learning Course for building AI-enabled Prediction Services from models and features
A six-module self-paced course teaching ML practitioners how to build end-to-end prediction services without managing infrastructure. It uses GitHub Actions for pipeline orchestration, Hopsworks for a feature store and model registry, and Gradio/Streamlit for UIs. The target is someone who can train a model in a notebook but has never shipped one to production.
The FTI (Feature, Training, Inference) pipeline separation is the right mental model and the course builds it incrementally across modules rather than dumping it on you at once. The credit card fraud example runs through multiple modules with real synthetic data, so you aren't constantly context-switching to new toy problems. GitHub Actions as the scheduler is a practical choice — students get CI/CD experience without needing to stand up Airflow or Prefect. The free-tier-only constraint (Hopsworks + GitHub Actions) means you can actually complete the course without a credit card.
Hard dependency on Hopsworks is a significant lock-in — if you want to understand what a feature store actually does under the hood, or use Feast/Tecton at work, you'll need to mentally translate everything. The course stops at XGBoost and sklearn; there's nothing on deep learning deployment, ONNX, or model serving latency concerns. Last push was September 2024 and the Orchest content (module 01) references a platform that shut down in 2023, which signals the material isn't being actively maintained. Module 06 README is nearly empty, so the real-time capstone module is the least documented part of the course.