JioStar’s chief architect on AI discovery at 500M-user scale

JioStar, which runs entertainment and live-sports platforms used by more than 500 million people, built a conversational assistant into JioHotstar using the OpenAI API. Viewers can ask in natural language for shows, match highlights or cricket statistics, across more than 15 languages, and the team says it built the experience in about eight weeks ahead of a major cricket tournament.

This comes from an OpenAI customer series, so it's a vendor showcase. But the lessons from chief architect Vijay Seshadri are unusually practical, and they apply far beyond streaming.

We'll be upfront that this is a promotional interview on OpenAI's own site, and it doesn't share outcome numbers like engagement or accuracy. So we're not grading the product's success here. What's worth our attention is how the team says it got to production, because that's where most enterprise AI projects stall.

The most useful line of thinking is that the team decided first what the system should not answer. Guardrails before features. For an open voice interface used by hundreds of millions, that's the right order, and it's the opposite of how many demos get built.

Then there's evaluation. Seshadri describes a seed dataset expanded by models, a maker-checker setup where one model answers and another judges, and domain experts like sports-magazine editors helping define what serious fans ask. He also raises the honest problem: who tests the judge? That question doesn't have a tidy answer, and we appreciate that he names it rather than waving it away. Models grading models can quietly share blind spots, and at this scale small errors become large ones.

The framing of accuracy, latency and cost as a constant three-way tradeoff is the real-world version of AI engineering. With more than 60% of users choosing voice, responses have to be fast and cheap as well as right. His regret, not investing earlier in monitoring model quality in production, is a lesson other teams should take seriously.

We're also glad he pushes back on vanity metrics, saying lines of code generated don't prove value. The flip side of his point is the risk: give an engineer a powerful tool without the judgment to validate production systems, and the risk multiplies along with the output. That's the honest version of the AI productivity story.

JioStar says it has supported as many as 72.5 million concurrent live-sports viewers. The conversation took place at OpenAI's Executive Summit in Singapore, where OpenAI's forward-deployed engineers were described as working alongside JioStar's small team.