Streaming Audio (Kafka / Confluent)

Inside OpenAI’s Streaming Backbone with Aravind Suresh | Ep. 24

In this episode, Aravind Suresh, head of OpenAI's real‑time infrastructure team, explains how the company built a highly reliable, scalable streaming backbone for products like ChatGPT using Kafka and Flink. He describes the challenges of scaling a streaming platform tenfold every six‑seven months, the need to hide infrastructure complexity from internal users, and the design choices—such as proxy‑based multi‑cluster Kafka and relaxed ordering guarantees—that enabled high availability and simplicity. The conversation highlights the importance of building simple, well‑abstracted systems that can evolve with rapid growth, and how OpenAI’s approach balances velocity with stability for mission‑critical AI workflows.