Any serious clinical AI product built for India probably ends up here.
In SF May 16–24. If you’re working on multimodal/audio LLMs for non-English markets or healthcare AI, would love to meet. @SarvamAI@GeminiApp
“Sir, patient ne kaha ki itching ho rahi hai for the last two weeks.”
A typical sentence in an Indian clinical consultation. Hindi, English, sometimes Marathi — code-switching per sentence, not per conversation.
Every AI scribe I’ve tested breaks on it.
Open weights haven’t kept pace. The only open audio-LLM for Indian code-switch is Shuka v1 — Aug 2024, built on Llama 3 8B, <100 hours of audio.
The strong stuff (Sarvam Audio, Gemini) is API-only. For fine-tuning on your own domain, there’s still a real gap.