Fox One: Video Content Personalization Powered by GenAI

*Copresenting with Fox.

When Fox introduced Fox One, its short-form video platform, we needed to deliver engaging recommendations at launch despite sparse, noisy engagement signals. Rather than waiting months for reliable data, we used GenAI to bootstrap personalization and guide early decision making under real-world constraints.In this talk, we describe how we designed a modular, production-ready personalization system on Databricks that blends GenAI with classical machine learning and evolves as data matures. We will cover how we used Gemini to extract structure from video content, leveraged Databricks Functions to infer early preference signals, built millisecond-level video retrieval on Databricks Lakebase, and developed offline evaluation using GenAI to provide directional insight when traditional A/B testing was unreliable. Attendees will learn how we weaved GenAI into classical recommendation systems on the Databricks lakehouse to ship personalization under data constraints.