Semantic IDs for Recsys with Chronon

Semantic IDs are emerging as a powerful technique for building modern recommendation systems, enabling models to reason over users, items, and content through compact representations. This talk is a technical deep dive into how semantic IDs work in recommender systems, why teams use them, and where they can improve retrieval, ranking, personalization, and cold-start performance.

We will also examine the technical challenges of implementing semantic ID–based systems in practice such as managing freshness, joining online and offline signals and avoiding training-serving skew.

Chronon is the open-source feature and embedding platform developed at Airbnb and Stripe, designed to help teams build, serve, and iterate on production ML systems. Attendees will learn how Chronon enables faster and safer experimentation with semantic IDs.