HealthMitra: An AI Voice Health Companion for India, Built in 10 Days
An AI health companion that answers — and places — phone calls in Hindi and English, built on LiveKit and Murf Falcon over ten days of shipping one feature at a time. By Siddharth Agarwal · Health Access track, 10-day voice-agent build · github.com/siddharthagarwal1983/sid-voice-for-bharat The problem, and who it's for Most healthcare-access tools assume a smartphone, a data plan, comfortable…
HealthMitra is an AI voice health companion for India, developed in just ten days as part of a voice-agent challenge. Its purpose is to assist callers who may face barriers to healthcare access, such as those who do not have smartphones, data plans, or reading comfort, or who may need to switch between Hindi and English during a conversation.
The AI voice companion aims to help callers determine the seriousness of their symptoms and whether they require home care, a visit to a Primary Health Centre (PHC), or an ambulance. It also provides information on eligibility for government schemes like PM-JAY (Ayushman Bharat) and assists ASHA community health workers with quick reference answers during their visits.
The system operates by opening every call in English and re-checking the caller's language on every turn, ensuring that the response is provided in the appropriate language. HealthMitra uses LiveKit for audio transport and four stages to convert sound into a response and back into sound: Deepgram Nova-3 for multi-language speech recognition, Gemini 3.5 Flash-Lite for reasoning and tool calls, Murf Falcon for text-to-speech (TTS) with an Indian voice tuned for code-mixing, and a SQLite database for caller memory and analytics.
The TTS stage is critical, as its latency directly impacts the caller's perception of responsiveness. HealthMitra addresses this by using Murf Falcon's 130ms time-to-first-audio benchmark, ensuring a smooth and efficient conversational experience for the caller.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
