JioStar built a 15-language chat assistant for JioHotstar in 8 weeks

JioStar's project began in October and November, with a major cricket tournament coming in February, leaving roughly two months to serve two very different audiences: entertainment viewers and cricket fans. An entertainment viewer might ask for a surprise on a Friday evening or describe a movie without knowing its title, while sports viewers want specifics such as fall of wickets, a series of sixes, or the match in which a player scored a certain number of runs. Cricket experts may want statistics about every ball. Bringing those discovery problems into one experience across multiple languages in a short period shaped the design.

The team's first decision was scope: because a voice or text interface lets users ask almost anything, they began by deciding what the system should not answer, putting guardrails first. For evaluation, they started with a seed dataset and used models to expand it with more variation, creating a maker-checker pattern in which one model helped determine how to answer a question while another acted as a validator with a competing objective. That raised the question of who tests the judge, making the validator itself an evaluation problem. Codex helped accelerate work such as generating eval cases and updating the system, but faster building did not remove the need for judgment and validation.

Seshadri's advice to executives moving from experiments to production systems centers on three points: a large gap between what technology can do and the reality inside an enterprise, a focus on outcomes rather than activity such as lines of code generated, and the human skill and agency required. He warned that without the judgment to validate production systems, risk multiplies alongside impact.