Who we are

We make machines India can talk to.

We build the full voice stack — our own models, our own infrastructure, down to the phone call — so any business can hold a real conversation with any customer, in their language, at human speed.

Why we’re doing this

India runs on conversations.

More than 1.3 billion people. One of the biggest markets on earth for conversation. Yet too little voice infrastructure is built around Indian languages, Indian prices and the way people here actually speak.

Here, people call to confirm an order, ask one more question, or hear that someone will take care of it. Reassurance is part of how business gets done. A conversation makes it personal.

That’s why we’re here. Building here. Hosting here. Running here. Voice Infra made in India, for the conversations India depends on.

The story, in six beats

  1. India is voice-first.

    The next billion users do not type — they talk. Business here closes on calls, not clicks.

  2. Conversations don’t scale.

    A human calling team is the most expensive, slowest-to-hire and hardest-to-train bottleneck in Indian business.

  3. Voice AI kept failing the trust bar.

    Slow, robotic, English-first, rented from abroad. People hang up on machines that do not listen.

  4. We built the whole stack ourselves.

    Our own voices, our own audio pipeline, our own speech recognition and language model, in Indic languages — judged on real calls with real customers, never on benchmarks. 900 ms median to first reply, because the person on the phone should not wait for our architecture. Speed is respect.

  5. It works today.

    Businesses run calling campaigns from ₹1 a minute that book real meetings. Developers use the same models through simple APIs. Four to five thousand minutes a day, in production.

  6. Where it goes.

    Every business in India able to talk to every customer — and eventually, a billion voice-first people talking to every service, through us.

Fifteen months, bootstrapped

  • Fifteen months in market since June 2025, with no outside money — every rupee earned went back into the stack.
  • The first invoice was ₹8.25 a minute. The floor today is ₹1: eight times down, and not by discounting. We removed the vendors instead.
  • Two paying customers on our own infrastructure — one running production campaigns end to end, one on the TTS API.

How we work

  • Ship by measurement, not by vibes. A release that sounds fine still gets rolled back if the numbers moved.
  • The deploy log is the git log. Nothing is hand-edited on a live box.
  • Latency is the first constraint, always. Recording, logging, billing and analytics are designed off the hot path.
  • Every rupee is accounted. The fleet controller carries the ledger and refuses to start a machine it cannot prove it can pay for.
  • Pilot-first. One brand, minimal components, live calls, then scale.
  • Indian-first, not India-also. Hinglish native to the engine, Devanagari rendered with the same care as Latin, rupee pricing, audio that never leaves the country.

A note from the founder

I left my job in Silicon Valley to work on voice full time. For nine months we fought the same fight as everyone else — making agents as human and as accurate as we could. Then the market handed us a reality check: at ₹5 a minute, voice AI does not sell in India.

So we went after the cost. The only honest way was to own the stack: our own GPUs, our own models, our own inference engine, our own SIP lines. That is what got it to ₹1 a minute.

What is behind this code is that stack, running live. Break it, push it, tell us what it gets wrong.

— Sneh Mehta, founder

Try the sandbox Read the journal
Try the sandbox