He almost failed computer science. Now Y Combinator and Z Fellows are backing this teen in Delhi

Samrath Singh Chadha was sitting in the middle of a Navy SEAL retreat in the Wisconsin wilderness – being pushed well beyond his limits – when the emails started arriving. Y Combinator, the legendary Silicon Valley accelerator that launched Airbnb, Dropbox and Stripe, wanted to know more about his startup idea. At 18 years old, the Bangalore-raised founder had apparently caught their attention. He just wasn't entirely sure why.

"We were sure we (him and his co-founder) had bombed the application," he says, laughing softly. He is 19 now, running on what sounds like very little sleep, speaking from Delhi where he studies mathematics – and, tentatively, philosophy – at Ashoka University, when he is not building artificial intelligence systems from scratch.

The story begins, as many good ones do, with failure. In eighth grade, Samrath received a D in his computer science exam. Mortified, he cracked open a textbook and began teaching himself C. Within weeks, what started as remedial self-study had become an obsession. He moved to Python, built a password manager at fourteen, entered hackathons, and somewhere in the fog of pandemic lockdowns discovered the Discord communities where a generation of young founders were quietly assembling their futures.

It was there he met his eventual co-founder, a boy a year younger than him who happened to live in a suburb of San Francisco. They had never met in person. They spent nights hacking away at projects that, by Samrath's own admission, "had no value, would rarely even go up on GitHub." But they were having the time of their lives.

The first proper venture came when he was thirteen – an internship placement platform for Indian high school students, which he named InternNova. It placed around fifty students, half in paid positions. It also taught him, brutally, about chasing payments from companies that never intended to pay and about the exhaustion of running something before you are ready. By fifteen, it was over.

GPT passion

By the time he was seventeen, the world had discovered ChatGPT – but Samrath had already spent months conversing with GPT-3, back when his friends assumed there had to be a human on the other end. He had found something that consumed him completely: the idea that you could compress the whole of human knowledge into a system that speaks and reasons. He begged approximately fifty Google engineers for cloud credits to play with these models. One replied. He was the global head of developer relations. Samrath left that call with a thousand dollars in credits and a conviction that this was where his future lay.

The summer after graduating high school, he flew to the United States for a family trip and ended up spending six weeks sleeping in his internet friend's house in the Bay Area. On the last day before flying home, the two of them submitted a Y Combinator application almost on a whim. "We weren't ready for it," he says. "It was like a practice run."

Two weeks later, the interview invitation arrived. Samrath joined the YC interview from a Starbucks car park en route to a Z Fellows retreat in Wisconsin (another coveted accelerator opportunity), and spent three days receiving increasingly probing follow-up questions while being put through physical hell at a camp in the woods. In the end, YC accepted their idea.

The financial terms of a Y Combinator deal are standard across every company in a batch. Founders receive $500,000 – $125,000 in exchange for a 7% equity stake, and a further $375,000 that converts into equity when the company raises its next funding round, at whatever valuation that round commands. For Samrath, a simultaneous acceptance into Z Fellows – a smaller, more intimate programme that he describes as perhaps the most intellectually formidable group he has encountered, operating in batches of around ten – added a further $100,000, bringing the total raised to $600,000 before he had written a single line of code in San Francisco.

Pivots and more pivots

After taking a semester-long break from college in Delhi and settling into an apartment in San Francisco, Samrath and his co-founder threw themselves into the YC batch's rhythm. The cohort of around 250 founders from 100 companies turned out to be the most densely talented room he had ever entered. “Brian Chesky, the co-founder of Airbnb, was present for the kickoff. It was so cool,” he says.

And while every conversation crackled, the product they had arrived with – a voice AI agent for website landing pages – was, he admits plainly, something neither of them particularly believed in. The pivoting began almost immediately.

They then built Mantle: an autonomous agent system that connected to a company's entire operational stack – CRM, calendar, payments, email, analytics – and could watch for real-world triggers and act on them without human instruction. Within roughly two and a half weeks of selling, they were generating $10,000 a month in revenue. Then, abruptly, they refunded every customer and shut it down. The problem wasn’t one they found terribly interesting.

"We told our customers we were pivoting away," he says.

What came next was lonelier. His co-founder stepped back. The company became, effectively, just Samrath. Rather than retreating, he threw himself into a new and considerably more technical problem. Coding agents – the AI systems that write and navigate software on a developer's behalf – spend the vast majority of their time not writing code at all, but simply searching through codebases trying to understand what is where. He is now training a small, highly specialised model, sub-one-billion parameters, to do that search work faster. The model maps a codebase as a semantic graph, understands the structural relationships between different parts of the code, and learns to navigate it the way a senior engineer might – intuitively, purposefully, without brute-force trawling. Early results on the SWE-bench evaluation – one of the standard measures of coding agent performance – show meaningful improvements on the hardest problems.

Samrath is aware that his trajectory runs against a tide of anxiety about what AI means for young people entering the workforce. He does not dismiss those concerns. "There will be a huge amount of displacement," he says. "That's kind of inevitable." But he is equally convinced that for those who learn to build with these tools, the demand is, for now, surging. What worries him more is the concentration of power over technology that will affect every person on the planet. "We can't have a few billionaires deciding for us," he says.

The application to Y Combinator, he adds before signing off, is free, takes about three hours, and does not require American citizenship. He recommends it even to those who do not intend to accept the offer, simply for the discipline of articulating who you are and what you are trying to build. It is, he suggests, the kind of exercise that has a way of surprising you.