Term · Topic
Feature Store
Feature store is a centralized system that manages, stores, and serves machine learning features for both training and inference, enabling consistent feature use, reuse, and governance across data science teams and production environments in enterprise machine learning platforms.
- 2 articles
- 2 companies
- Most covered: Hopsworks, Redis
What is Feature Store?
A feature store is a centralized data system that manages, stores, and serves Machine Learning (ML) features for training and inference in a consistent and reproducible way across models and environments.
A feature store functions as a dedicated layer in the ML stack that organizes feature data used as inputs to models. It standardizes feature definitions, computation logic, storage formats, and access methods so that training and production systems use the same features.
Core characteristics include feature ingestion from batch and streaming sources, feature transformation and validation, storage in online and offline repositories, and low-latency serving for real-time inference. It also supports metadata management such as feature schemas, provenance, and versioning.
Enterprise Usage and Architectural Context
Enterprises deploy feature stores within data and ML platforms to support model development, testing, deployment, and monitoring across teams. They typically System Integration Testing (SIT) between data lakes or data warehouses and model training or serving infrastructure, including Machine Learning Operations (MLOps) pipelines.
In this context, the feature store coordinates with orchestration, workflow, and monitoring tools to enforce consistent feature computation and access patterns. It can integrate with identity and access management systems and logging services to support governance and audit requirements.
Related or Adjacent Technologies
Feature stores relate to data warehouses, data lakes, and operational data stores but focus specifically on ML feature lifecycle management. They complement model registries, experiment tracking systems, and MLOps platforms that manage models and deployment workflows.
They also interact with stream processing frameworks and Extract, Transform, Load (ETL) or Extract, Load, Transform (ELT) pipelines that compute or update features from raw data. In some enterprise architectures, feature store capabilities appear as part of a broader data platform rather than as a separate product.
Business and Operational Significance
For enterprises, a feature store supports reuse of curated features across multiple models and teams, which can reduce duplicate engineering work and inconsistencies in feature logic. It enables more repeatable training and evaluation because historical feature views remain accessible for analysis.
Operationally, it supports more consistent behavior between offline training and online inference by serving the same features in both contexts. It also provides a structured point to apply governance controls over which data features models can access and under what policies.
Our definition.
Companies in this area
Listed by how often coverage ties them to this term.
