📄️ Feature stores
Why training and serving compute the same feature two different ways, and how Feast, a registry and point-in-time joins fix it. 5 hours, applied projects, a 40-question exam and a verifiable certificate.
📄️ 1. Training-serving skew
Module 1 of the Feature Stores premium course: two implementations of the same feature, silent performance drift, a worked example on card fraud, and what a store promises to fix.
📄️ 2. Anatomy of a feature store
Module 2 of the Feature Stores premium course: registry, entities, feature views, sources, offline and online stores, and the feature server placed in an MLOps architecture.
📄️ 3. Offline and online store
Module 3 of the Feature Stores premium course: full history for training versus latest value at low latency, typical technologies, and how the two stores stay consistent.
📄️ 4. Defining and versioning features
Module 4 of the Feature Stores premium course: declarative feature definitions, types, TTL, ownership, and evolving a definition without breaking existing models.
📄️ 5. Point-in-time correct joins
Module 5 of the Feature Stores premium course: how a naive join leaks the future into training, what an as-of join actually does, and the effect on metrics.
📄️ 6. Materialization and freshness
Module 6 of the Feature Stores premium course: batch materialization to the online store, cadence, streaming features, measured freshness, and cost.
📄️ 7. Feast in practice
Module 7 of the Feature Stores premium course: a feature repository, apply, get_historical_features, materialize, get_online_features, the feature server, and the limits of Feast.
📄️ 8. Features shared across teams
Module 8 of the Feature Stores premium course: discovery, documentation, reuse between fraud and marketing, ownership and governance, the cost of duplication avoided.
📄️ 9. Monitoring feature quality
Module 9 of the Feature Stores premium course: missing values, distribution, freshness, alerts upstream of the model, and the link with drift from MLOps.
📄️ 10. Project: feature store for scoring
Module 10 of the Feature Stores premium course: end-to-end pipeline on card fraud, measuring skew before and after, serving latency, and what it would have taken to do without.
📄️ Recap and exam
Complete recap of the Feature Stores premium course: skew, anatomy, stores, definitions, point-in-time joins, materialization, Feast, sharing, monitoring, project — then the 40-question exam.