Feast-workshop is a collection of educational modules in Jupyter Notebook format designed to teach users how to use Feast, an open-source feature store.
The workshop addresses the practical challenge of managing machine learning features at scale. Feast solves three core problems: enabling feature reuse across teams, performing point-in-time-correct data joins for training data generation, and operationalizing features for online inference while preventing training-serving skew. The tool provides both low-latency online stores for real-time predictions and offline stores for batch scoring. However, Feast deliberately does not handle data pipeline orchestration, data quality monitoring, or experiment management; instead it integrates with adjacent tools like dbt, Airflow, and Spark to fit into existing workflows.
Teams should adopt this workshop if they need hands-on learning about feature store concepts and best practices through practical examples. The modules build sequentially, so they work best for developers new to Feast who want to understand both foundational concepts and common use cases. The workshop requires a substantial local setup including Python, Docker, Jupyter support, and depending on the module, Terraform and either AWS or GCP credentials. Developers should be aware that M1 Macbook development is untested and Windows development requires WSL.
The project maintains structured, progressive learning materials organized into modules that build on each other, with explicit prerequisites documented for each stage. The workshop acknowledges platform-specific limitations and provides guidance for developers working in constrained environments. The materials reference both the main Feast documentation and specialized tutorials like the Great Expectations integration guide, positioning the workshop as part of a broader learning ecosystem rather than a standalone resource.
