Ask most people what artificial intelligence is made of and they will name a chatbot. Jananan Mithrakumar, who goes by Jana, spent four years working on the answer underneath that: the silicon. He now runs a New York company that has raised 2.25 million dollars from Silicon Valley investors and built a subscription business earning around 1.5 million USD a year, from software used by more than 300,000 people.

Jana left Colombo for Stanford University in 2015, one of a very small number of Sri Lankans admitted that year. He arrived from Elizabeth Moir School with five A* grades at A Level, a result almost no student in the country matches in a given year, along with ten A* grades at IGCSE and an Edexcel World Prize for ICT, the highest mark in the world in that subject. He is quick to put all of it in its place. "The results opened a door, and that is all a result can do. What actually mattered was what happened after I walked through it, which was ending up in rooms where I was the least experienced person there and staying long enough to stop being that."

He took a bachelor’s and then a master’s degree in Electrical Engineering, choosing the narrow seam where artificial intelligence meets the hardware it runs on. From his second year, he worked in Stanford’s robotics laboratory under Professor Oussama Khatib, one of the discipline’s founding figures, on a robotic interface able to operate inside an MRI scanner while a subject’s brain was being imaged. The research was published at IROS and IEEE World Haptics, two of the field’s principal international conferences. His closest collaborator over those two years was a senior researcher in the laboratory named Samir Menon.

You left Colombo for Stanford in 2015. What did you think you were going to do?
Honestly, I had no plan beyond getting there. I knew I liked building things and I knew I was interested in how computers actually work at the physical level, which is not a fashionable interest at eighteen. I did not know that this was a career, or that there was an industry that would pay for it. The plan came later. What I had at the time was curiosity and a reasonably good set of results, and only one of those turned out to matter.

You were in Stanford’s robotics laboratory from your second year. How does an undergraduate get into a lab like that?
By being useful and being persistent, in that order. Nobody in that laboratory owed me anything. I asked to help, I did the unglamorous work, and eventually I was given real problems. Professor Khatib is one of the founders of the field, and I was a second-year student from Colombo. The gap was enormous. But laboratories always have more work than hands, and if you are willing to do the part nobody wants, you get to stay in the room.

You then went to intern at Dexterity, the robotics company your collaborator in the lab had founded.
Almost everything that has happened to me came out of working closely with somebody who was better than I was at the time. Samir and I spent two years on hard problems together. When he left to start Dexterity, that relationship already existed. It was not a network in the way people mean the word. It was two years of shared work, and then an internship at the company he built. I would say that is the more reliable version.

At SambaNova you worked on power circuitry. Explain that for a reader who has never thought about a chip.
A processor has a clock, and every tick of that clock is one round of calculation. If you make the clock tick faster, the chip does more work every second, and that number is what the entire artificial intelligence industry is judged on. The limit is electrical. Push the frequency up and the power draw and the heat rise with it, and at some point the chip becomes unstable and fails. My work was on the circuitry that governs that trade. You are trying to buy speed without buying instability.

Why leave a company worth 11 billion dollars?
Because it was going to be fine without me. That was rather the point. Large companies are full of enormously talented people, and the honest truth is that any individual contribution disappears into the whole. I wanted to work on something where it mattered whether I turned up. That is a selfish reason rather than a noble one, but it is the real one.

Y Combinator takes fewer than one applicant in a hundred. How do you prepare for that?
You do not, really. Twenty-four thousand applications and two hundred and twenty-nine places. You cannot polish your way through odds like that, and the people who try tend to produce something that sounds rehearsed. What you can do is be extremely clear about what you are building, who it is for, and why you specifically are the person to build it. Clarity travels. Polish does not.

Your first product did not work. What happened?
It worked; it just did not grow fast enough. We were building software for manufacturers, automating the paperwork I had watched consume ten hours a week of engineers’ time at every company I had worked at. Real customers, real problem, insufficient speed. So we pivoted, which means we kept the company, the money and the team and changed what we were building. Falling in love with your first idea is the most expensive mistake a founder can make. The company is the asset. The idea is a guess you are allowed to revise.

Why move from software for manufacturers to something aimed at everybody?
Partly growth, but mostly because I sat down and worked out what I actually wanted to spend a decade on. Selling to manufacturers is honest, useful work and it serves a small, specific group extremely well. I kept coming back to how many people it could ever reach. I wanted to build something my cousins could download. Not a minority of specialists, but millions of people who would never call themselves technical. Once I was clear about that, the decision made itself.

Smoothspeak has more than 300,000 users and around 1.5 million USD in annual
recurring revenue. What did you get right?
We stopped guessing and started listening. The product is built for conversations that happen on
a screen, and the thing that made it work was explaining the reasoning rather than handing somebody a line to paste. People do not want to send a message they did not mean. Beyond that, we priced properly for markets outside the United States instead of converting dollar prices and hoping. Willingness to pay in South Asia is real. Very few companies do the work to price for it.

A second app, Elo, is close to launch. What is it?
Microlearning and brain games, powered by artificial intelligence. Short lessons and short exercises you can do in a few minutes, on a phone, standing in a queue. It comes from the same instinct as Smoothspeak. The interesting question about this technology is not what it can produce on your behalf. It is whether you are any better at the thing afterwards. A model that writes your message for you leaves you exactly where you started. I am far more interested in the version where you improve.

You have raised 2.25 million dollars in total. What is that money actually for?
Early on, you have no revenue at all, so somebody has to fund the attempt. That is what it is for, and in our case it also funded a pivot, which is an expensive thing to survive. But raising is not an achievement in itself, and I think our industry has confused the two rather badly. It is a loan against a promise, and the people who wrote those cheques expect something built. Revenue is the only number that tells you whether anyone actually wanted the thing.

Sri Lankan students often assume the ceiling is lower than it is. What would you tell them?
That there is nothing in what I did that a student sitting in Colombo cannot do. The people I studied alongside at Stanford were not working from some secret advantage. They had simply been told earlier that this was available to them, and they applied without agonising over whether they deserved to. If there is one thing I would change about my own path, it is that I would have told people sooner.

Can a genuinely global company be built from Sri Lanka?
The engineering can absolutely be done from Colombo, and increasingly it is being done there. The part that takes deliberate work is distribution, because distribution runs on relationships and trust, and those take time to build wherever you are in the world. That is the thing that requires more effort from a distance. It is not impossible, and Sri Lankans working abroad are a real asset here. Most of us are far more reachable than people assume. Write to us. We answer.

And to a parent reading this whose child has just come out with good results?
Let them aim higher than feels reasonable. I applied to Stanford half expecting nothing to come of it, and that application is the single highest-return thing I have ever done. A no from an institution eight thousand miles away costs you nothing at all. Not applying is the only version where you are certain of the answer. And once they get there, the marks stop counting almost immediately. Nobody is coming to find you because your marks were good. You have to go and stand in the room.