JioStar, an entertainment and live-sports platform serving over 500 million consumers, built a conversational discovery assistant into JioHotstar using OpenAI's API. The system allows viewers to speak naturally to find movies, shows, historical highlights, live-match moments, and cricket statistics across more than 15 languages. Vijay Seshadri, Chief Architect at JioStar, led the effort from concept to production in approximately eight weeks, ahead of a major cricket tournament in February.
The technical challenge centered on balancing three competing constraints: accuracy, latency, and cost. More than 60% of JioStar's users prefer voice input, making response speed and economic viability at scale critical. The team framed content discovery as fundamentally different across three user segments—entertainment viewers seeking casual recommendations, sports fans looking for specific moments, and cricket experts wanting detailed statistics—and unified all three into a single conversational experience. Domain experts, including sports-magazine editors, were essential to seed evaluation datasets; the team used a maker-checker pattern where one model generated answers while another validated them.
For commerce practitioners scaling AI in production, Seshadri emphasized the importance of early investment in model-efficacy monitoring beyond traditional infrastructure metrics. He highlighted the gap between frontier-model capabilities and enterprise reality, and cautioned against mistaking visible activity for meaningful outcomes. The project underscores that shipping AI at consumer scale requires clear scope definition, guardrails, rigorous evaluation frameworks, and cross-functional teams working alongside AI platform providers.