RL environments and training data to make AI superhuman at physical engineering.

Fourier builds RL environments and training data to make AI superhuman at physical engineering, plus the software and integrations that bring these models to engineers.

LLMs are now superhuman at mathematics and coding, primarily due to the shift to RL environments with automated, robust verifiable rewards. Every field with computable, verifiable outcomes is going to follow suit.

Physical engineering, from nuclear fusion reactors to quadcopter drones, is mostly done on a computer. Engineers design in CAD, test the design in physics simulators, and iterate in a loop. That is an RL environment, and the simulators are the verifiable reward. Fourier is a complete CAD/CAE platform, built on mature open-source CAD and simulation tools. Anything an engineer can do in the UI, an AI agent can do through the API.

Engineering a complex system means optimizing a Pareto frontier in thousands of dimensions. We use that structure to spawn a curriculum of RLVR tasks from a single CAD design, from improving individual metrics while keeping all others constant, up to rebuilding the entire assembly and beating the reference’s Pareto frontier. Multiplied across hundreds of thousands of real designs, that produces mountains of training data.

We don’t see AI’s ability to engineer being capped at what humans have done. Truly general purpose robotics, commercially viable nuclear fusion, cheap access to space, and much more that right now we can only dream of.

To get involved or invest, contact us at adamkutak@gmail.com.

RL environments and training data to make AI superhuman at physical engineering.

You’re on the list.