Aayam Bansal sold a patented AI device at 17, built a helpline for tens of thousands, and left college weeks in. Now, at 18, he is trying to compress the timeline of scientific discovery itself.
The rejection email arrived like they always do - polite, final, and completely unhelpful about what came next. Aayam Bansal and his research partner had spent months on a machine learning paper. The Association for Computational Linguistics passed on it. Two teenagers who had already published at some of the biggest venues in artificial intelligence looked at the process that had just dismissed their work and asked a different question than most people ask. Not "how do we fix the paper?" but "how do we fix the thing that judges the paper?"
That flip - from improving the submission to rebuilding the system - is the whole personality of Bansal's career in one moment. He is 18. By the time most people his age are picking a college major, he had already patented an AI device, sold it, built government-scale infrastructure, and co-authored research read by working scientists. The pattern repeats: hit a wall, and instead of climbing over it, take the wall apart to see how it was built.
Bansal grew up in India with a habit of building things that solved problems in front of him. During the COVID-19 pandemic, that instinct turned into helpline infrastructure that served more than 50,000 people a day - not a prototype or a class project, but a running system carrying real load during a crisis. It is the kind of thing that would anchor an entire resume for most people. For him it was an early chapter.
At 17, he built, patented, and sold an AI orthopaedic device. The company was called aisock, and the sale went through for a five-figure sum. Think about the sequence there: identify a physical-world problem, apply machine learning to it, protect the idea legally, and find a buyer - all before graduating from a stage of life most people spend memorizing for exams.
The research came in parallel. Bansal worked across MIT CSAIL, Carnegie Mellon, the National University of Singapore, and Harvard - a map of labs on three continents. His Google Scholar record shows a spread of curiosity that refuses to sit in one lane: temporal encoding for energy time-series, linguistic patterns in hip-hop lyrics, fairness in algorithmic lending, drift detection in sentiment models, and machine learning applied to Formula 1 race performance. It reads less like a specialist's portfolio and more like a mind that wanders into a room, gets interested, and publishes on the way out.
It was in these labs that he met Ishaan Gangwani, a competitive programmer ranked in the top fraction of a percent globally. The two published together at NeurIPS, ICML, and AAAI workshops. Long before there was a company, there was a working partnership - two people who already knew how the other one thought under deadline pressure.
Synthetic Sciences - originally launched under the name InkVell - is built on a single, uncomfortable observation. Over roughly 24 months, software engineering got dramatically faster because of AI. Scientific research did not. The literature review still takes weeks. The experiments still stall on GPU queues. The paper still takes months. Bansal's bet is that the same acceleration can come to science, and that whoever builds the tooling for it is standing at the front of something enormous.
The product is an AI co-scientist. A researcher hands over a task, and the system runs the loop: synthesizing literature, generating hypotheses, designing experiments, executing code, managing GPU jobs, and producing publication-ready output. It targets machine learning research and computational biology. In biology mode, the company reports 92% on BixBench Verified - a benchmark number that made investors look twice.
Investors agreed the timing was right. The company raised roughly $1.5 million, including $500,000 from Y Combinator, with pre-program backing from the a16z Scout Fund, Pioneer Fund, and Amplo VC. It landed in YC's Winter 2026 batch as one of the youngest founding teams to secure funding at that level before even starting the program.
Bansal was admitted to the University of Illinois Urbana-Champaign to study computer science. He did not finish his first semester. When Y Combinator came in, the math changed - not the math of tuition versus salary, but the math of time. A window like this does not wait for a diploma. So he closed the laptop on his coursework and opened it on the company.
The day-to-day since has been a blur of roles that would normally belong to a dozen different people. He describes it plainly.
He does not pretend it is comfortable. "The constant context-switching is exhausting," he has said. But he frames the intensity as the point, not the price: "At this age, you're learning all of it in real time, which is intense, but it also forces you to level up very quickly." It is the same instinct from the rejection email - discomfort is just a signal that the wall is worth taking apart.
For all the seriousness of the mission, Bansal keeps a sense of play visible. His public bio lists his research fellowships and his conference publications next to a detail he clearly enjoys: two God of War platinum trophies. It is a small thing, but it says something honest. This is a person who finishes what he starts, whether that is a hundred-hour game or a research agenda, and who does not feel the need to hide the game to look serious about the science.
The through-line across all of it - the sock, the helpline, the papers, the co-scientist - is impatience with slowness. He watched research move at the speed of committees and queues and deadlines, and decided that speed was a solvable engineering problem rather than a law of nature. Whether he is right is the open question of the next few years. But the willingness to bet his teens on it is already answered.
The story is still early. The company is two people and a benchmark and a lot of ambition. But the shape of it is familiar to anyone who has watched Bansal work: find the slow thing, refuse to accept it is slow, and start building the faster version before anyone gives you permission to.