Abhilash Chowdhary spent a decade putting autonomous robots on real sidewalks. Then he turned a bootstrapped side project into Crustdata - a real-time data layer for AI agents that just raised a $6M seed with a one-person sales team.
The robots came first. Before the API calls and the seed round and the phrase "search engine for agents," Abhilash Chowdhary was a robotics engineer watching small autonomous machines roll down real sidewalks with someone's dinner inside. His teams at Serve Robotics built one of the largest food-delivery robot fleets in the world and pushed past 10,000 deliveries. He rode that company from an idea to a listing on the NASDAQ. Then he left to build something you will never see on a sidewalk.
Crustdata does not move atoms. It moves information - real-time data about companies, people, and events, delivered through simple APIs to the software that now reads the web on our behalf. The pitch is unglamorous and enormous at the same time: if AI agents are going to make decisions, someone has to feed them facts that are current and structured. Chowdhary decided that someone would be him.
The resume reads like a tour of hard engineering problems. He grew up in India, studied computer science at IIIT Hyderabad, and moved to the United States in 2015 for a master's degree. His first American job was at Thinknum, building a data intelligence platform for public-market investors - the seed of an idea he would carry for years. Before that world fully clicked, he stacked up systems experience: petabyte-scale data pipelines at Yahoo, large-scale products at Adobe used by millions of people.
Then came robots. He was a founding robotics engineer at Robotwits, a Carnegie Mellon spin-off later acquired by Waymo, working on self-driving technology. He led robotics at Postmates X, building the autonomous delivery platform that Uber eventually absorbed. And he ran robotics engineering at Serve Robotics from its earliest days through its public offering. Each step was a lesson in shipping complex systems into the messy real world, where a sensor glitch is not a bug ticket but a robot stuck at a crosswalk.
Crustdata did not begin as a company. It began as a nagging idea. Back at Thinknum, Chowdhary had seen how valuable clean, real-time company data could be for investors. He spent roughly two years turning the concept over in his head - at one point imagining a version built for the Indian market - before writing much code. In startup mythology, two years of thinking sounds like procrastination. For him, it was conviction being assembled slowly.
The project launched publicly in March 2022 while he still held a full-time engineering job. The stack was deliberately boring and cheap: Python on the backend, React on the front, Postgres underneath, and less than $100 a month in software subscriptions kept afloat partly by AWS credits from Y Combinator's Startup School. Co-founder Manmohit Grewal, who had crossed over from law into software engineering, came on to handle the frontend while Chowdhary owned the backend.
"If we had focused on our customers rather than our competitors, we could have saved ourselves at least 6 months."Abhilash Chowdhary
The early growth story is a small masterclass in being wrong productively. Chowdhary's first instinct was content marketing and chasing features that competitors already had. His first serious sales push targeted institutional investors: more than 1,000 cold emails went out, around 20 people replied, and not a single one converted. It would have been easy to read that as proof the idea was dead.
Instead he changed the audience, not the effort. He aimed at private investors rather than institutions, sent fewer than 100 emails, and landed his first paying customer inside a month. From there he built a method he could repeat: hypothesize the customer's specific pain, write subject lines under seven words, and test pitch variations across cohorts of a hundred people at a time. One campaign hit a 3% conversion rate - a number that would make most outbound teams jealous.
Somewhere in that grind, the market moved toward him. AI agents started needing exactly what Crustdata was quietly good at: fresh, structured facts about the business world. The company reframed itself around that shift. Chowdhary describes the ambition plainly - index the important data on the web and hand it back in a form software can actually use.
"Think: a search engine for agents, where the output isn't '10 blue links,' but entity-linked, structured, fresh context from the web."Abhilash Chowdhary
The traction followed the thesis. Revenue climbed from roughly $770,000 to more than $4 million in seven months, achieved with zero outbound marketing and a sales team of one. Crustdata reached profitability by dogfooding its own product rather than burning cash on go-to-market theater. The customer list became its own proof point: Y Combinator uses it for founder discovery, Dharmesh Shah's agent.ai runs on it, and Ryan Reynolds' MNTN is in the mix.
In its Fall 2024 batch, Y Combinator took Crustdata in, and in early 2026 the company closed a $6M seed round led by YC and A Capital, with General Catalyst, SV Angel, and Phosphor Capital along for the ride. The two main co-founders had originally met through YC's co-founder matching tool in 2023 - one in San Francisco, one in New York - which is a fittingly data-driven way to start a data company.
"AI agents are only as intelligent as their data sources."Abhilash Chowdhary
What comes through in his own telling is a bias toward pragmatism over polish. He self-funded, kept overhead near zero, and treated discomfort as a signal to move rather than retreat. His regret is not that he was too scrappy but that he waited to talk to customers - the six months he thinks he lost building against competitors instead of listening to buyers.
He keeps a line from Phil Knight's memoir Shoe Dog close: "The cowards never started, and the weak died along the way." It is the kind of quote founders either print on a wall or actually live by. For someone who spent two years thinking before launching, then rode two hardware companies through acquisitions and an IPO before bootstrapping a third, the second reading fits better.
The through-line across robots and data is the same instinct: take something hard to get from the physical or digital world and make it usable at scale. Delivery robots turned "someone has to drive across town" into a machine rolling down the block. Crustdata is trying to turn "the web is a mess of pages" into clean context an agent can act on. Different medium, same job.
Whether the agent economy grows into the trillion-dollar thing its believers expect is still an open question. But if it does, it will run on data that is fresh, structured, and trustworthy - and Abhilash Chowdhary has spent his whole career learning how to build exactly that kind of plumbing, first for robots, now for the software reading the world.