Problem
Most AI products assume typing in English. In India, that locks out millions. Voice is faster, more natural, and language-aware — but wiring STT → LLM → TTS with low latency is hard.
Voice AI agent · Sarvam AI · LiveKit · Featured
VANI — multilingual voice agent
Text-first AI excludes people who prefer to speak. VANI is a voice-first assistant built on Sarvam AI — talk in your language, ask anything, get a spoken answer. Hiring a Sarvam AI / LiveKit voice developer? This is the demo.
Solution
VANI uses Sarvam AI for multilingual speech-to-text and text-to-speech, with a LiveKit realtime pipeline: user speaks → Sarvam STT → LLM → Sarvam TTS → playback. FastAPI backend handles memory and logic. No typing needed.
How it works
LiveKit transports audio. Sarvam handles Indian languages. LLM reasons. TTS speaks back. The pieces are modular — swap models, add RAG, or connect to your APIs.
Stack
Sarvam AI · LiveKit · Python · LLMs · FastAPI · Voice pipeline
FAQ — VANI
Can you build this for my product?
Yes. I ship voice agents for support, onboarding, and internal tools — prototype in 5-10 days, then harden for production.
Which languages?
Whatever Sarvam AI supports — Hindi, English and more Indian languages, extensible to others.