About the job

YOUR MISSION

A software engineer today uses Claude Code or Cursor. A mechanical engineer designing the next jet engine still searches Excel sheets and asks whoever has been around longest.

We are building the AI co-engineer for physical products: one that works natively in 3D and learns from real engineering decisions. We are in production with major European manufacturers, with real revenue and customers who tell us when we are slow.

The problems are open ones. Getting an agent to reason usefully about geometry, making 3D search work on real customer part libraries, and figuring out what the interface even looks like when an engineer asks why a bracket failed. There is research to borrow from and very little product to copy.

Senior or staff level depending on experience.

What You'll Work On

You will own large problems end to end, across the stack. We cannot say exactly what that looks like in month three, but here is what the last two years have actually involved.

• A multi-agent orchestration SDK, built from scratch. Nothing off the shelf handled what we needed: long-horizon execution that runs for hours and survives restarts.
• CAD conversion as a real pipeline. It used to be manual and brittle. Turning it into production infrastructure meant writing a Python wrapper around a C++ SDK and making it reliable enough to sit under everything else.
• Migrating the application off Electron and Webpack onto SSR React with Vite. That architecture is what the whole product now sits on.
• 
  

What is coming:

• 3D-to-3D search across real customer part libraries, not benchmark datasets
• Constraint solvers and optimization, and getting an agent to use one sensibly
• Agents that hold a useful model of a physical object, which is still an open problem
• Hybrid cloud deployments for manufacturers who cannot send their data anywhere
• 
  

We work in Python and TypeScript, with multi-agent orchestration and AI integrations across several providers.

What We Care About

• You have shipped production systems across frontend and backend, and can point at them
• Your CV reads as a set of problems, not a stack
• You use AI heavily to go faster, but you built things the hard way first
• You are fine when the problem is not yet defined
• You deliver end to end, and you are still there when it needs maintaining