I built an AI commentary engine that takes ball-by-ball cricket data and produces broadcast-style commentary with audio — as if you're tuning into a live match on radio.
Demo: https://cricvox.in (India vs Pakistan, T20 World Cup 2026, Hindi commentary with audio)
How it works: Match data flows through a state engine (tracking score, stats, momentum) → a logic engine that classifies each delivery and picks only the narratively relevant context → GPT-4.1 generates commentary → ElevenLabs TTS synthesizes audio with emotion tags ([excited], [tense], [dramatic tone]).
Key design choice: the LLM never sees how a ball was bowled or what shot was played — only bare facts (runs, wickets, match state). This prevents hallucinating specifics. Commentary focuses on narrative, momentum, and stats — things it can reason about truthfully.
I built an AI commentary engine that takes ball-by-ball cricket data and produces broadcast-style commentary with audio — as if you're tuning into a live match on radio.
Demo: https://cricvox.in (India vs Pakistan, T20 World Cup 2026, Hindi commentary with audio)
How it works: Match data flows through a state engine (tracking score, stats, momentum) → a logic engine that classifies each delivery and picks only the narratively relevant context → GPT-4.1 generates commentary → ElevenLabs TTS synthesizes audio with emotion tags ([excited], [tense], [dramatic tone]).
Key design choice: the LLM never sees how a ball was bowled or what shot was played — only bare facts (runs, wickets, match state). This prevents hallucinating specifics. Commentary focuses on narrative, momentum, and stats — things it can reason about truthfully.
This is interesting. Any fine tuning on existing commentary?