Refract Systems is building the open, AI-native version of the cyber-physical engineering stack — the things Ansys and Siemens do, rebuilt open-source and directly with AI. VirtMCU is live. Here's the rest of it, and who I need.
I treat AI-written code the way chip designers treat transistors: nobody hand-places the billions of them. Increasingly I don't hand-write the if and while statements either — the AI lays them down. Which means the entire game is verification: proving the system does what it must. That's the lens I build through, and it's how one person ships like a team — a continuous fleet of AI agents refactoring, reviewing, and testing 24/7.
VirtMCU — register-level microcontroller emulation over MCP, built for AI agents. Live and monetized.
Co-simulating firmware with physics engines — VirtMCU already wired to MuJoCo and NVIDIA Omniverse.
CAD, materials, and a full design-to-manufacturing pipeline, closed back into the simulated environment.
The end state is what I think of as a Tony Stark lab: one engineer with the automated leverage to design, simulate, verify, and manufacture a real cyber-physical system entirely from a terminal. Every stage is a monetizable, self-sustaining product — we scale on cash flow and independence, not on raising.
My own depth is software, simulation, and AI orchestration — so I have that covered. I'm explicitly not looking for another me. I want one, maybe two, peers with PhD-level depth in a complementary layer. If you're already co-simulating firmware against physics, I'd genuinely like to compare notes on how you're connecting it.
Electromagnetics, power management, signal integrity. (There's a constructive-attenuation idea from my PhD I want to revive.)
Closed-loop control, estimation, and the applied math that makes a digital twin track reality.
Correctness as the real product — the way EDA verifies silicon. If you're building RL-driven verification (automated test generation, fuzzing loops, formal-ish guarantees), we should talk.
Parametric modeling and the geometry kernel that the simulation and manufacturing loop is built on.
Material models and properties that let a twin predict the physical part, not just its shape.
Process simulation and the path from a verified design to a part that actually gets made.
No comfortable corporate day job. You're full-time on your own thing, building and shipping. You don't need to be convinced, managed, or babysat.
You can own a whole system solo — from raw idea to production — but you've outgrown the solo-developer ego and want the multiplier of a high-density crew.
You don't just use AI tools; you orchestrate fleets of agents and ship what used to take a team. I've had impressive people want in — the gap was never talent, it was whether they'd actually change how they work.
You're physically here. I've learned the hard way that even a commute from SF breaks the whiteboard-and-walk loop this stage needs.
To be clear on what this isn't: no PMs, no managers, no kingdom-building, no process-heavy bloat. Peers, not subordinates.
Spin up a VirtMCU session, make it do something that would've needed a bench full of hardware, and send me the link plus three lines on why it mattered. That tells me more than any CV.