Redis launches Feature Form for production machine learning

Redis has launched Redis Feature Form, a managed feature store for production machine learning, further extending its presence in the machine learning software stack.

The launch follows Redis's acquisition of Featureform and brings feature definition, orchestration, versioning and serving into a single platform. The system is designed to help machine learning teams move models from testing into production while reducing operational work and inconsistencies between training data and live inference.

A recent Deloitte report found that only a quarter of businesses have successfully moved artificial intelligence pilots into production. Redis is targeting that gap with software designed to manage the full lifecycle of features, the structured data inputs used to train and run machine learning models.

The platform combines offline training workflows with online serving, an area that has often required separate systems and custom integration. By bringing those functions together, teams can define features once and use them across both model training and inference.

Platform Changes

New additions include support for batch and streaming pipelines, including backfills, incremental updates and tiling. Redis has also introduced workspaces for multi-team use, allowing organisations to separate data, authentication and observability by workspace.

It has added job planning, impact analysis, split materialisations and queue-based job management to give teams more visibility into changes before they affect production systems. Another change is atomic management of graph-level updates, rather than versioning individual resources separately, to simplify rollback and change tracking.

Security updates include workspace-level access controls, API key pairs, audit logs, secret-provider updates, mTLS and encrypted internal transport. Redis has also reduced deployment complexity with a two-service deployment model and rebuilt the dashboard to support workspace and provider configuration through the interface.

Production Focus

Feature stores have become an important part of machine learning infrastructure because they help organisations manage the data features used by models in development and production. Problems can arise when the data used to train a model differs from the data available at deployment, leading to performance issues or model drift.

Redis is positioning the platform for production use cases such as fraud detection, credit and risk scoring, personalisation and recommendation systems, where delays or inconsistencies in feature data can have direct commercial effects. It is also targeting platform teams that have built internal pipelines and want a more governed path to self-service machine learning work.

The move broadens Redis's position in the market. The company has traditionally been used as a low-latency data layer in larger machine learning architectures, especially for online serving. With Feature Form, Redis is seeking a larger role in defining, governing, and orchestrating features across an organisation.

That puts Redis in closer competition with vendors offering feature stores and broader machine learning operations tools. It also reflects wider demand for software that can connect experimental artificial intelligence work to stable production systems without extensive custom engineering.

Simba Khadder, Head of Context Engine at Redis, said the product is intended to reduce complexity for machine learning teams: "Feature Form helps ML teams move features from definition to production with less glue code, less drift between training and serving, and less operational overhead." 

"It is building toward a future where Redis supports both modern AI systems and long-lived production ML workloads from the same data platform foundation," Khadder said.