Independent software studio · Punjab, India
We start things because we want to find out. We finish them because half-built answers aren't answers.
01 — premise
Most software gets built because someone asked for it. Ours gets built because something wouldn't leave us alone — a question about whether a thing was possible, or possible cheaply, or possible for people who don't speak English to a computer.
Curiosity is the easy half. The hard half is being able to finish: to take the interesting question all the way down to a thing that runs on a Tuesday, for a stranger, without you in the room. That second half is what makes a studio rather than a hobby.
02 — method
The name is the process. It's an acrostic because we needed to remember it, not because it looked good on a slide.
Before writing anything, find the actual constraint. Usually it isn't the one in the brief. On a voice agent everyone optimises the language model; the bill is really speech synthesis, charged per character. Getting that ordering right decided the whole architecture.
Build the smallest thing that can be wrong in public, then let it be wrong. We keep a harness that runs the whole system without the expensive parts attached, so a bad idea costs minutes instead of invoices.
Finish it. Handle the ugly cases — the person who interrupts, the answer machine, the network that drops mid-sentence. The gap between a demo and a product is entirely made of these.
03 — evidence
Two shipped, one in the workshop. Each started as a question we couldn't put down.
“Why does everything you learn from a video evaporate in a week?”
Learn it. Build it. Defend it.
A tutor that lives inside your AI assistant. It adapts to you, makes you build something real, then refuses to pass you until you can defend the work to a stakeholder who pushes back.
“Why does one short video need a camera, an editor and a timeline?”
Type a topic. Get a quiz reel.
Animated quiz reels with hand-drawn mascots, in your brand colours, written by an AI that knows the format. No camera, no editor, no timeline — three clicks and it renders.
“Can a phone call in Hindi cost less than a cup of chai?”
Real-time conversation, priced in paise.
A bot that phones a lead, qualifies them in Hindi or Hinglish, and hangs up knowing whether they're worth a salesperson's time. The interesting part was never the conversation — it was the economics.
not yet public
04 — craft
Anyone can describe a system. These are the details that only show up if you actually built one.
Cost is a design material
Speech synthesis bills per character, so scripted lines are rendered
once and replayed forever. That single decision moved a call from
₹2.92 to ₹1.4 a minute — measured on real
calls, not modelled in a spreadsheet.
Hindi isn't a translation layer
Indian callers switch language mid-sentence. So the system transcribes Hinglish as Hinglish rather than forcing it into one script — and the voice on the other end was chosen by listening, not by reading a spec sheet.
We test the thing we ship
Every product has a way to exercise it end to end without the expensive parts connected. It's how you find the bug that only appears when a real person interrupts at the wrong moment.
Numbers, or it didn't happen
If we claim a thing is fast or cheap, there's a measurement behind it and a way for you to re-run it. Confidence without instrumentation is just decoration.
05 — currently
The honest state of the workshop. Some of these become products. Most won't.
We take on a small number of build projects a year, mostly where the problem is genuinely unsolved.