What is a feature store, and what problem does it solve in ML systems?

A feature store is a centralized system for defining, computing, storing, and serving ML features. It typically has two parts: an offline store (a data warehouse or lake) that holds historical feature values for training, and an online store (a low-latency key-value store like Redis or DynamoDB) that serves fresh features for real-time inference. Its main value is consistency: the same feature definition and transformation logic produce values for both training and serving, which reduces training-serving skew. It also enables feature reuse across teams and models, point-in-time correct joins to avoid label leakage, versioning, and monitoring. Popular implementations include Feast, Tecton, and cloud-native offerings like Vertex AI and SageMaker Feature Store.