Real-Time Features, AI Search, and Agentic Similarities - Building Next-Generation ML Infrastructure

Explore why traditional feature stores might be the wrong abstraction for modern AI/ML infrastructure in this podcast episode featuring Varant Zanoyan, Co-founder & CEO at Zipline AI, and Nikhil Simha Raprolu, Co-founder & CTO at Zipline AI. Discover how Chronon ditched "store-first" thinking to focus on compute, orchestration, and real-time correctness, drawing from battle-tested experience at Airbnb and Stripe. Learn about the evolution from feature stores to feature platforms, understand the challenges with open-source solutions like Feast and Feather, and examine how Zipline integrates with Apache Iceberg for scalable data management. Delve into the complexities of building agent systems at scale, explore the relationship between features and embeddings in modern ML workflows, and gain insights into why real-time ML, embeddings, and AI agents often feel painful to implement. The discussion covers practical architectural decisions, the origins of Chronon as an open-source project, and how next-generation AI/ML infrastructure platforms can streamline data pipelines, model deployment, observability, and governance to accelerate enterprise AI development.