API World 2023 + AI DevWorld 2023

Nick Acosta, Tecton, Developer Advocate

ML is increasingly used in production to power a broad range of product experiences, including fraud detection, recommender systems, real-time pricing, search, and risk analytics. However, these real-time ML systems present unique data challenges. They must incorporate batch, streaming, and real-time data to use the freshest data available in an organization. They also need to provide historical data with point-in-time correctness for model training and low-latency serving at scale for online inference. Without the right tooling in place, it can take weeks or even months to build the complicated pipelines to support these requirements.

In this talk, I will present the role of a modern feature platform to address these challenges so teams can build real-time ML models quickly and reliably. A feature platform enables teams to design, deploy, serve, centralize, and manage ML features that incorporate batch, streaming, and real-time data.

I’ll show how a feature platform integrates with your existing data and model serving infrastructure to:
Develop and manage features as code, enabling fast iteration using DevOps engineering best practices
Unify batch, streaming, and real-time transformations using an optimized feature transformation framework
Automate the entire feature lifecycle, from versioning features and processing feature data, to serving features for training and online inference, and logging feature values
Generate accurate training datasets
Serve fresh features online for real-time inference with enterprise service levels
Monitor data pipelines and operational service levels
Share and re-use features across teams

A modern feature platform is an essential tool in the MLOps stack. It allows teams to build better features and more accurate models, get to production faster and accelerate the speed of iteration, and deploy new services with enterprise-grade reliability, security, and performance.

Join me to learn how feature platforms provide a new, modern data foundation to solve some of the hardest challenges in building ML-powered product experiences.