A better base for AI

With many enterprises struggling to move AI and machine learning projects into production, Feature Form is a new set of capabilities designed to help Redis customers better manage the models that ae essential to AI development.

Feature Form is meaningful because it moves Redis beyond fast serving and further into the day-to-day management of features. Operationalizing features across teams and environments is one of the core production ML challenges for enterprises at scale.

Building underlying machine learning models is no longer the primary problem preventing development of AI and machine learning tools, according to Redis. Instead, problems arise after models are built when attempts are made to deploy models across teams and environments.

Model drift that makes a model's training data stale, machine learning pipelines that break down in production due to changes in their underlying data or deployment in a new environment, and gaps in governance are among the obstacles that enterprises encounter.

Feature Form is designed to help Redis customers keep model training and model serving synchronized once models are in production across different enterprise environments and deployed to fuel AI and predictive analytics initiatives. Some of those customers, meanwhile, were struggling to get machine learning features into production and provided the impetus for Redis' acquisition of Featureform and subsequent introduction of Feature Form, according to Simba Khadder, founder and CEO of Featureform and now AI product lead at Redis.

"Our largest customers were telling us about the operational challenges of getting ML features into production -- real-time data pipelines, versioning and lineage, consistency between training and serving," he said. "At the same time, we were watching teams build homegrown solutions on top of Redis to handle those problems. … The feedback confirmed we should go further."

Feature Form includes the following:

• Unified batch file and streaming data pipelines to reduce the amount of work required of developers and engineers to customize machine learning model pipelines.
• Workspaces for organizations with multi-tenant use of single Redis instances so that teams can isolate their work.
• Fine-grained job control to provide teams with greater visibility into changes in their data before they write data to other systems or unseen changes accidentally affect production systems.
• Improved access control and security measures.
• A new deployment model aimed at reducing complexity while still enabling advanced patterns.
• A redesigned user interface that supports all new workflows enabled by Feature Form.
• Atomic directed acrylic graph (DAG) -- visual representations of data models and their connection to one another -- updates to make change history easier to view.

Unified streaming data and batch file pipelines are perhaps the most valuable of the new features because it reduces some of the custom engineering work required of machine learning teams, which can struggle to keep separate pipelines aligned, according to Pratt.

However, he noted that although Feature Form is beneficial for Redis users, its capabilities are not unique among database vendors, with AWS, Databricks, Google Cloud and Snowflake providing similar tools that address machine learning workloads. Redis nevertheless has the potential to distinguish itself by bringing improved governance and orchestration capabilities into a database platform already known for its low-latency serving and real-time data workloads, Pratt continued.

"Feature Form gives Redis a credible way to stand out by making the platform more complete for production ML," Pratt said.

Catanzano, who highlighted unified pipelines and atomic DAG updates as Feature Form's most valuable capabilities, similarly noted that although other database vendors provide similar capabilities, there are ways Redis' new feature set for AI and machine learning is unique.

"Redis Feature Form differentiates itself by integrating directly with Redis' real-time data platform," he said. "This combination of sub-millisecond performance and enterprise-grade feature management is unique, positioning Redis as a leader in production ML environments."

Looking ahead

With Feature Form now in preview, Redis' primary product development focus is on building database capabilities that will enable customers to access contextually relevant information for their agentic AI development initiatives, according to Khadder.

Agents require relevant data to perform as intended. Often, however, that data is difficult to discover and deliver.

"We're focused on one core problem: making context usable by agents," Khadder said. "Most companies don't have an agent problem. They have a context problem. ... Our focus is on the foundational pieces that solve these problems."

Specifically, Redis is building a context engine that unifies structured data and unstructured data along with memory in a real-time context layer that agents can call on.

To further improve its machine learning capabilities, Catanzano recommended that Redis add support for more prebuilt models and integrate its database with popular third-party machine learning frameworks such as TensorFlow and PyTorch.

"Additionally, focusing on industry-specific solutions, such as tailored feature stores for healthcare or finance, could attract new customers while deepening its value for existing ones," he said.

Pratt, meanwhile, suggested that Redis deepen observability capabilities related to Feature Form to provide users greater visibility into whether features are fresh, stable and performing as expected before they affect model performance.

"A strong next step for Redis would be deeper feature observability, giving customers more confidence as they scale production ML," he said.

Eric Avidon is a senior news writer for Informa TechTarget and a journalist with more than three decades of experience. He covers analytics and data management.