Two macro problems in hardware design
Kyrall is a Munich, VC-backed startup focused on design automation for hardware. Osama’s framing: tools are fragmented, and work inside each tool is still manual - click-heavy CAD plus weak connectivity upstream and downstream.
Results from sizing tools are used as input, alongside requirements, specifications, existing CAD, and mission definitions. Kyrall is not building another requirements or sizing product. From that set it generates parametric models that can continue into analysis and manufacturing.
“More circular” design - and why a mesh is not enough
A customer line he quoted sets the ambition: design should be more circular. Change a calculation or requirement, and CAD geometry and simulation should update; simulation should flow back into the design. Engineers become orchestrators of that loop.
The technical bar is where demos usually fail. A usable model is not any 3D mesh. It has to be B-rep, with sketch constraints and assembly mates, so it opens in CAD as if a designer built the feature tree. The same bar applies downstream: topology-optimization STL back to parametric CAD; manufacturing method changes that reshape geometry without redrawing every step by hand.
Three use cases and three adoption gaps
In delivery: conceptual design (thousands of early configurations from sizing XML), asset generation (drawing or photo toward a parametric part with feature tree), and STL to parametric CAD (mesh from simulation, scan, or legacy sources).
Adoption is still slow, he argued, until three gaps close. The first hurdle is the lack of models that run locally, on ordinary machines inside the company. Fine-tunes on company CAD and design habits are a further step. After that: reliability via automatic evals, and deep integration into the tools the company already runs, rather than a product that sits on top of the stack.
Getting legacy CAD into that loop is harder than training another model. Geometry on its own is not enough - the model also needs design intent, why the part looks that way.
