A spreadsheet full of analysts that never sleep
Hebbia's flagship product is called Matrix. Feed it ten thousand pages - a data room, a decade of filings, a pile of contracts - and it answers in a grid, row by row, showing its work like a junior analyst who actually cites sources.
That is the bet George Sivulka has been making since 2020: that the most valuable people in finance, law, and consulting spend most of their day on tasks no human should have to do. Reading. Cross-referencing. Copying a number from page 412 into a cell. Hebbia's job is to take that drudgery off their desks and hand it to AI agents that work the way a person would, only across millions of documents at once.
By mid-2024 the thesis had teeth. Hebbia raised a $130 million Series B at roughly a $700 million valuation - on $13 million of revenue, and profitable, which in the cash-incinerating world of generative AI reads almost like a typo. Andreessen Horowitz led. Index Ventures, Google Ventures, and Peter Thiel were already in. Revenue had grown 15x in eighteen months. Roughly a third of all asset managers were running the product.
"You have the most intelligent people coming out of Stanford and all of these amazing schools, and the majority of them go to jobs where they're just rendered automatons."
- George Sivulka, on the brain drain that started HebbiaThe kid who dialed NASA
As a teenager he picked up the phone and cold-called NASA. It worked. He landed an internship at NASA Goddard, which tells you most of what you need to know about how he operates - the assumption that the door is unlocked if you just try the handle enough times.
He carried that to Stanford and finished a mathematics degree in about two and a half years, reportedly with a 4.0. Then, not satisfied, he went to war with the registrar. "I petitioned the administration 10 times until they let me take more units than was possible," he has said, "more units than almost anyone has ever taken as an undergraduate." He slid into a master's in applied physics as a Threshold Venture Fellow, then a PhD in electrical engineering, working at the seam where machine learning meets theoretical neuroscience.
The company's name is a tell. Hebbia comes from Hebbian learning - the neuroscience principle that neurons which fire together wire together. He was naming a startup after the wiring of the brain while still studying it.
From a $500 closet to Thiel's living room
The PhD was fully funded. He left it anyway. The frustration was specific: he watched the smartest people he knew graduate into jobs that turned them into copy-paste machines, and decided that was a problem worth quitting school over.
The romantic-startup math did not add up at first. Before the term sheets, he was living in what he has described as a roughly $500-a-month closet near campus, struggling to make rent. Then, by 22, he was spending hours with Peter Thiel, arguing about artificial intelligence and philosophy. Thiel - who reportedly called him a wunderkind - wrote one of Hebbia's first checks.
He built the thing as a solo founder. No co-founder, no business pedigree, a team stocked largely with fellow Stanford researchers. He was also early to something that later got a fashionable acronym: Hebbia was among the first to productionize Retrieval Augmented Generation - RAG - inside real enterprises, before most of the industry had a name for stuffing a model's context with the right documents at the right time.
Try the handle
A teenage cold call to NASA became an internship at Goddard. The same nerve later got him more course credits than nearly any Stanford undergrad in history.
The closet years
Roughly $500 a month, rent a struggle - and then a standing seat across from one of tech's most contrarian investors.
Solo by design
No co-founder, no MBA. Just a roster of researchers and a conviction that knowledge work was ripe for rewiring.
Early to RAG
Hebbia was productionizing retrieval-augmented generation in the enterprise before the acronym went mainstream.
"I'm excited for a world of unbound progress - one where AI agents contribute more to global GDP than every human employee."
- George Sivulka, on what he is actually building towardAgent employees
Plenty of people sell chatbots. Sivulka's pitch is bigger and more uncomfortable. Chatbots, he argues, handle the easy questions and choke on the hard ones - the multi-step, cross-referenced, source-everything work that defines an actual professional. So Hebbia's roadmap is not assistance. It is the "agent employee": software that takes a task end to end, the way you would, only faster and at a scale no human can match.
He has said the quiet part on the record - that bots will make up most of the economy within a decade, that agents will eventually contribute more to global GDP than people. It is the kind of line that sounds like hype until you remember he built a profitable company by being right about a smaller version of it first.
The customers are not hobbyists. Asset managers, investment banks, law firms, pharma - Centerview Partners, Charlesbank, Fenwick & West among the named ones. These are organizations that bill by the hour and have every reason to be skeptical of a 20-something with no MBA telling them his software can read their data room faster than their associates can. They signed anyway.
Watch: George Sivulka on 20VC - The Future of Foundation Models →