ML Infrastructure
7.0k 2026-04-13
feast-dev/feast
An open-source feature store that streamlines the management and serving of features for AI/ML models, ensuring consistency between training and inference.
Core Features
Ensures consistent feature availability for training and serving via offline and online stores.
Prevents data leakage by generating point-in-time correct feature sets.
Decouples ML from data infrastructure with a unified data access layer.
Manages historical data for scale-out batch scoring or model training.
Powers real-time prediction with a low-latency online store.
Quick Start
pip install feastDetailed Introduction
Feast is an open-source feature store designed to simplify the lifecycle of machine learning features. It enables ML platform teams to efficiently manage, store, and serve features for both model training and real-time inference. By providing a unified data access layer, Feast ensures feature consistency, prevents data leakage through point-in-time correctness, and abstracts away underlying data infrastructure complexities. This allows data scientists to focus on feature engineering and model development, accelerating the deployment of robust ML systems.