JioStar chief architect Vijay Seshadri is running AI recommendation and discovery systems for 500 million users, one of the largest deployments of its kind anywhere. The platform uses OpenAI models in production, not in pilot, and Seshadri's team has built evaluation frameworks and guardrails specifically designed to hold up at that scale.
The piece is worth reading for the specifics: how Seshadri's team designs evals before shipping features, what guardrail failures actually look like at 500 million users, and the gap between what works in a demo and what survives contact with real traffic. These are operational details, not theory, and they are rare in public writing about AI deployment.
If you are building anything that touches content discovery or recommendation at scale, this is a primary source. The lessons on taking AI from prototype to production are concrete, and Seshadri names the failure modes his team actually encountered, which is more useful than any benchmark.
[READ ORIGINAL →]