Scaling Real-Time ML Infrastructure at Coinbase

From fraud detection to personalization, Coinbase relies on real-time machine learning to power high-impact, user-facing systems — and deliver measurable results. Their real-time ML platform has contributed tens of millions of dollars to the business impact by enabling models that detect fraud and drive personalized recommendations at scale.

But serving low-latency predictions across these use cases requires infrastructure that can support dynamic data, fast iteration, and production-grade performance.

In this moderated Q&A, Coinbase Senior Software Engineer Joseph McAllister joins us to share how their ML Platform team built a flexible, real-time feature platform that powers deep learning models trained on live user behavior.

Joseph will talk about how Coinbase:

• Serves predictions with sub-second freshness and <100ms latency
• Scales batch and streaming pipelines using Tecton and Databricks
• Enables ML engineers to ship features to production without wrangling infra
• Built a reusable framework for sequence features that power deep learning models trained directly on user behavior

Whether you're focused on fraud, personalization or infrastructure for modern ML systems, this session offers practical insights from one of the most advanced ML platforms in fintech.