Hirsh Jain

Every enterprise has the same graveyard. It's where the AI pilots go. The slick demo that wowed the boardroom, the prototype that never touched a real customer, the proof-of-concept that proved nothing except how far legacy infrastructure can be from frontier ambition.

Hirsh Jain built Percepta to raid that graveyard. The pitch is almost rude in its simplicity: most AI vendors sell a tool and walk away; most consultants sell a slide deck and bill by the hour. Percepta sends its own engineers, researchers, and product managers to live inside your organization and carry the work from prototype all the way to production. It is a General Catalyst Transformation Company - meaning the venture firm didn't just write a check, it helped stand the thing up - and Jain runs it as co-founder and CEO out of New York.

The targets are not the easy ones. Percepta points itself at healthcare, financial services, manufacturing, and government - the industries that keep society running and that, not coincidentally, are the hardest places on earth to deploy anything new. Banks want fraud detection and a customer experience that doesn't feel like 2009. Manufacturers want their supply chains to stop surprising them. Governments want licensing and permitting that doesn't require a fax machine. Hospitals want clinical workflows that give time back to clinicians. Percepta's bet is that you can't buy any of that off a shelf. You have to embed.

To do it, Jain wired in the two partners that matter most for serious enterprise AI: Anthropic for the models and AWS for the cloud. Then he went and got the people.

The framing he chose at launch was telling. Percepta, he wrote when he finally introduced it, was about “ensuring that AI is a force for elevating and future-proofing the industries that matter” - not the consumer apps that grab headlines, but the slow, regulated, mission-critical machinery underneath. The post landed with more than a thousand reactions and a couple hundred comments, and ended with a hiring call: Percepta was recruiting in New York and across Europe, and he wanted people to apply. A founder's first public act is often a pitch to customers. His was a pitch to talent.

That instinct - that in this business the constraint is people, not product - runs through everything Percepta does. The company organizes its work around what it calls three transformation pillars: advancing the intelligence layer, building the data infrastructure beneath it, and driving the workforce adoption that most projects forget until it's too late. The first two are where the headlines live. The third is where AI initiatives quietly go to die, and it's the one Percepta refuses to treat as an afterthought.

Plenty of founders build a deck and call it a team. Jain co-founded Percepta with Hemant Taneja - not a fellow operator looking for a side bet, but the CEO of General Catalyst itself, the firm that has spent years placing some of the earliest chips on AI. When the person who runs your investor decides to put his own name on the cap table next to yours, that is a particular kind of signal.

The founding bench reads less like a startup and more like an all-star draft. Thomas Mathew. Athul Paul Jacob, out of MIT and Meta's FAIR research lab. Radha Jain. Michael Rochlin. And Costis Daskalakis - the MIT theoretical computer scientist whose work on game theory and computation is the sort of thing that gets named-checked in graduate courses. The wider team pulls from Palantir, MIT, Meta FAIR, Google, and Citadel. This is the part Jain seems proudest of, and the part that would soon land him in court.

Because if your business model is “deploy elite people into the hardest enterprises,” then your real product is the people. And the people are where the fight started.

The team's reported pedigree - Palantir, MIT, Meta FAIR, Google, Citadel - is not decoration. Each name is a shorthand for a different hard discipline: production-grade deployment, foundational research, scale engineering, and the kind of quantitative rigor that hedge funds pay fortunes to keep. Stacked together, they describe a company built to operate at the seam between research and reality, which is precisely the seam where most enterprises get stuck. It is also, not coincidentally, the kind of roster a much larger former employer might notice walking out the door.

Before he was anyone's defendant, Jain was the person enterprises called when the problem was too big to outsource and too important to fumble. He spent seven years at Palantir Technologies and rose to senior vice president, running the healthcare and civilian-government business - which is to say, he spent his prime years selling and deploying serious software into exactly the institutions Percepta now targets. He left in August 2024.

Rewind further and the pattern is a deliberate tour of the places that don't tolerate hand-waving. He studied computer science and mathematics at Harvard. He worked at Jane Street, the high-frequency trading firm where being wrong is measured in milliseconds and millions. He did time at Google. He worked inside the U.S. Department of Defense. Somewhere along the way he became a published machine-learning author. Each stop has the same flavor: high stakes, low patience for theater, real consequences when the code is wrong.

It is a strange and specific collection - finance, search, the Pentagon, big-data analytics - and it explains the shape of Percepta better than any mission statement could. Jain isn't trying to make AI cute. He's trying to make it survive contact with institutions that can't afford for it to fail.

There's a through-line worth naming. At Palantir, the entire premise was that you can't just hand a regulated enterprise a piece of software and wish it luck - you embed engineers, you sit with the data, you stay until the deployment is real. That model made Palantir one of the most valuable companies in the sector. Percepta is, in many ways, an argument that the same model can be unbundled from any single vendor's stack and pointed at the frontier - at Anthropic-class models and AWS-scale infrastructure - on behalf of whoever needs it. Whether that argument is a fair extension of an idea or an unfair appropriation of one is, of course, exactly what a court in Manhattan was asked to decide.