2025-11-05 –, Green Language: English
Production ML failures often stem from one overlooked issue: features that work perfectly in development break during inference. Through hands-on demonstration, this session shows how to eliminate feature drift using Feast's Python-based open source architecture. Learn to build reliable feature pipelines that maintain consistency across training and serving environments, ensuring your models perform as expected when deployed at scale.
Feature inconsistency is the silent killer of production ML systems. While data scientists perfect models in notebooks, real-world predictions fail because features differ between training and serving.
Feast (feast.dev), the rapidly growing open-source feature store, solves this by creating a bridge between data engineering and machine learning.
This session explores Feast's architecture that solves key ML challenges: feature consistency, point-in-time correctness, and low-latency serving. Through a churn prediction example, attendees will see how Feast's registry and store design handles both historical training data and real-time predictions.
Participants will learn to implement a production-grade feature management system that scales with their ML operations.
Previous knowledge expected
I'm the AI/ML Engineering Team Lead at Voyantis, where I act as a translator between data science, backend engineering, and business goals. My day-to-day involves overseeing our AI agent initiatives and building the shared infrastructure that powers them. I'm a big believer in finding the simplest path to a solution that works. I recently led our shift to a SQL-based ML architecture, which cut costs by 80% and onboarded customers 10x faster. I share these lessons in Towards Data Science and practice yoga - because debugging pipelines and holding a pose both require patience and balance.