The San Francisco company teaching machines - and airmen - to read the room. Photos in, photorealistic 3D training worlds out, with an AI instructor standing by.
San Francisco, California. Fourteen people. One large idea: that the next data an AI learns from won't be scraped off the internet - it will be captured from the world itself.
Somewhere on an Air Mobility Wing flightline, an airman pulls on a headset and steps into a maintenance bay that does not physically exist. The panel in front of them is photoreal, down to the scuffed fasteners. A voice - patient, tireless, not human - walks them through the procedure, watches their hands, and corrects a mistake before it becomes expensive.
That room, and the instructor inside it, are made by Schemata. The company calls what it builds "spatial intelligence," which is a tidy phrase for an untidy ambition: to give software an understanding of physical space, and to let people practice real-world work inside it. The trick is that Schemata doesn't build these environments by hand, frame by expensive frame. It captures them - from ordinary images and physical assets - and reconstructs them into interactive 3D worlds. The company's own numbers put it at roughly ten times faster and up to eighty-five percent cheaper than the traditional way of making training content, which historically involved film crews, 3D artists, and a great deal of patience.
It is a small company with an unreasonably large thesis. And for now, some of the most demanding customers on earth - the U.S. Navy's sea systems command, multiple Air Force units, a global food company, an oilfield services giant - are betting it's right.
Spatial Intelligence is the next iteration of machine intelligence. We will be able to effectively transition from internet-data ML training to world-data ML training.
Most "AI for training" turns out to be a chatbot bolted onto a slideshow. Schemata rebuilt the room the training happens in - then made the room teach. Here's how the pieces fit.
Photorealistic virtual training built from real physical assets and reality-captured data. Learners practice procedures, inspections, and maintenance in a faithful digital twin of the actual thing.
A built-in AI coach that explains procedures, monitors a learner's actions, gives real-time feedback, and adapts to each individual. An infinite supply of patient first tries.
Edge-native AI guidance that runs in secure, disconnected, or classified environments - where the internet doesn't reach and the stakes are highest.
Real-time dashboards that track progress, completion, and skill gaps - turning "did they get it?" from a gut feeling into a measurement.
Schemata's early roster reads like a list of places where "learn by doing" is either impossibly expensive or genuinely dangerous. That's the point. When the cost of a mistake is a grounded aircraft or a shut-in well, an unlimited number of photorealistic rehearsals starts to look less like a nice-to-have and more like insurance.
It's opening the door for so many opportunities for our airmen.
It sounds like the setup to a joke. It's actually a fairly deliberate assembly of complementary strengths: leadership and hard problems, computer-vision research and design, and the industry knowledge to sell into Washington.
Former USMC Infantry and Information Operations Officer. Mechanical engineer with two NASA fellowships. Stanford MBA plus an MS in Symbolic Systems. Once briefed the Secretary of the Navy.
Computer-graphics researcher out of Stanford's Virtual Human Interaction Lab. Gates Scholar with degrees in Symbolic Systems and Computer Science. The reality-capture brain.
Leads business development and customer operations. Brings DC industry knowledge and startup sales experience - the bridge between a demo and a defense contract.
Engineers and operators from Stanford, Berkeley, the military, and government start building 3D-capture tooling under the name Traverse.
The company becomes Schemata and widens its ambition from 3D content creation to spatial intelligence - the interface between AI and the physical world.
A seed round led by Owl Ventures, with a16z Speedrun, Alumni Ventures, Anorak Ventures, Time Zero Capital, and Plug and Play. Defense and enterprise pilots go public; the DoD is already testing the platform.
Product demos and interviews live on Schemata's own channels. Start here.
The airman pulls off the headset. The jet is still sitting there, untouched, exactly as it was. But something has changed: the procedure is now in muscle memory, the mistakes were made in a room where mistakes cost nothing, and the AI instructor logged every step for whoever signs off on readiness. The first time the wrench actually turns, it isn't really the first time.
That's the quiet reframing Schemata is after. Not a flashier simulator, not another headset cult - just the stubborn insistence that people should get to practice the real thing before it counts, and that machines should learn from the world the same way we do. Whether spatial intelligence becomes the next platform is still an open question. But a company of fourteen has already convinced the Navy, the Air Force, and a few Fortune 500 names to find out. The jet waits. The mechanic is ready. Nobody had to break anything to get there.