A flat photo, a 3D world, and a coach who never sleeps
Start with the thing on the screen. You hand Schemata a stack of ordinary photographs - a turbine, an engine bay, a control room - and it hands back a photorealistic 3D version you can walk through. Then it adds something stranger: an AI instructor living inside the space, one that answers your questions out loud, points at the right valve, and grades your work as you go. That is the product James Brown gets up to build every morning, and it is the clearest window into how he thinks. He is less interested in the robot than in the room the robot lives in.
Brown is the co-founder and CEO of Schemata, a San Francisco company of fourteen people with an unusually specific thesis: that the next leap in machine intelligence is spatial. Not text. Not images. Space. His phrase for it is blunt - "Spatial Intelligence is the next iteration of machine intelligence" - and the company's whole stack is built to back it up. Schemata uses 3D reality capture to rebuild real places, then layers what it calls semantic data on top, so an AI agent does not just see pixels but understands that this is a fuel line and that is an emergency shutoff.
The pitch lands because the alternative is so painful. High-fidelity simulators have always been the privilege of elite units and deep-pocketed programs - expensive, slow, hand-built. Schemata's claim is that it can produce training modules roughly ten times faster and up to eighty-five percent cheaper than the old way. The customer list reflects the ambition: NAVSEA, the 97th Air Mobility Wing, the 11th LRS, AFWERX, alongside enterprise names like Nestle, Nabors, and Northeastern University. Defense, energy, infrastructure - the unglamorous places where one wrong move on a real machine is genuinely dangerous.
Mechanically, the platform splits into a handful of jobs. There is the immersive 3D layer, photorealistic environments rebuilt from real physical assets. There is the AI instructor, a coach that takes voice or text, identifies objects in the scene, and walks a trainee through a procedure step by step. There are classroom tools - knowledge transfer from human experts, instructor analytics, performance tracking - and field operations, where the same training can be pushed to the edge on secure networks for point-of-need guidance. The throughline is that the expert's knowledge gets captured once and then redeployed endlessly, which is exactly the problem every industrial operator with an aging, retiring workforce is quietly panicking about.
By harnessing generative AI's reasoning capabilities, we can parse complex 3D data and gain deeper insights into real-world environments.
First in the family, last to count himself out
Before Stanford and seed rounds, there was Nebraska. Brown was the first person in his family to earn a college degree, graduating from the University of Nebraska-Lincoln in 2014 with a mechanical engineering degree and a robotics minor. Two NASA fellowships came along the way, the kind of line on a resume that most people would build an entire identity around. He treated it as a starting point.
Then he did the thing that defines the next decade: he joined the Marine Corps. Five years as an infantry officer, including a deployment to the Middle East with the 15th Marine Expeditionary Unit. A Poets&Quants feature on his path captured him third from the left during a live-fire exercise in the Mojave Desert - a detail that tells you more than any title. After the infantry he became an information operations officer in the Marine Information Group, supporting Indo-Pacific operations for about two and a half years and, at one point, briefing the Secretary of the Navy on programs he managed.
That is the resume of someone who could have stopped applying for things. Instead he applied to Stanford, fully expecting a rejection. "It's easy to count yourself out even before you begin," he later said. He got in - with a scholarship and the Botha Chan Innovation Fellowship, worth $75,000. His reaction was less triumph than disbelief: "I didn't think in my wildest dreams that I'd be sitting at the GSB today." He stayed an extra year for an MS in Symbolic Systems and did research in Stanford's Virtual Human Interaction Lab, the place where the Schemata idea quietly began to take shape.
It's easy to count yourself out even before you begin.
Out of stealth, into the deep end
Schemata co-founded with Huy Nguyen, now CTO, and the division of labor is honest about it. "I have experience building and leading teams, and engineering solutions to complex problems," Brown has said, crediting his co-founders for computer-vision research, design, defense-world fluency, and startup sales. The company joined a16z's Speedrun accelerator in cohort 003 and spent its early life heads-down, taking on government contracts before most people knew the name.
In May 2025 it stepped into the light. The seed round - $5 million, oversubscribed - was led by Owl Ventures with a16z speedrun, Alumni Ventures, Anorak Ventures, Time Zero Capital, and Plug and Play along for the ride. The detail that makes investors lean forward was tucked underneath: $3.4 million in Department of Defense contracts already booked in roughly eight months. Revenue first, fanfare second. For a defense-tech startup, that order matters.
Nguyen frames the technology as a beginning, not a destination: "We're connecting AI to rich 3D environments with embedded semantic data. Virtual training is just the beginning." Brown's framing is bigger still. The goal is open-vocabulary 3D scene understanding for agents - AI that can reason about any physical space it is dropped into, the way a language model reasons about any sentence. Training is the wedge. Spatial intelligence is the company.
In his own words
Spatial Intelligence is the next iteration of machine intelligence.
I have experience building and leading teams, and engineering solutions to complex problems.
I didn't think in my wildest dreams that I'd be sitting at the GSB today.
By harnessing generative AI's reasoning capabilities, we can parse complex 3D data.
There is a tidy logic to the whole arc once you stack it up. An engineer who learned how machines fail. A Marine who learned how training keeps people alive when machines fail. A founder who decided the gap between those two facts was a company. Schemata sells the simulator that the infantry version of James Brown would have wanted - high fidelity, fast to build, with a patient coach inside who never gets tired of explaining the same procedure twice.
He has not forgotten the door he almost did not walk through, either. Brown mentors prospective military applicants through Stanford's Veterans Club, helping other service members navigate admissions odds that look longer than they are. He calls it paying it forward. It reads more like someone closing the loop on his own best decision.
What is striking about the way Brown talks is how little of it is hype. The numbers do the bragging - $5 million in, $3.4 million already earned, ten times faster, eighty-five percent cheaper - and he spends his words on the thesis underneath. Generative AI, in his telling, is not a chatbot novelty but a reasoning engine that can finally be pointed at messy three-dimensional data and made to extract meaning from it. Get that right and the training product is only the first thing you can sell. Inspection, maintenance, robotics, autonomy - everything downstream needs an agent that understands the room it is standing in. That is the long bet, and the order he runs it in is deliberate: revenue first, then a round, then the bigger story.
For now the company is fourteen people in San Francisco, heads down, shipping. The founders' complementary strengths - Brown on team-building and engineering, his co-founders on computer vision, design, defense fluency, and sales - read like a deliberately assembled unit rather than a lucky collision of dorm-mates. Which, given where the CEO comes from, is probably the point. He spent a decade learning that the difference between a good outcome and a catastrophic one often comes down to whether the person at the controls was trained on something close enough to the real thing. Schemata is his answer to that, built in software, sold by the simulator.