OsamaAtwi

@amypretzel I think most people are approaching this topic wrong. Expecting AI to fully solve functional design overnight is like standing in 1910 and waiting for a car to hit 200 km/h to call it useful. It will get there. But there is already enormous value in what can be achieved today

text-to-CAD today generates geometry - but manufacturing needs: - tolerances (how much can dimensions vary?) - assembly relationships (how does this part fit with others?) - process constraints (can this actually be machined/molded/printed?) - material specs - surface finishes

What if you could generate hundreds of robotic gripper variants in minutes? Kyrall integrates directly with Onshape; describe the change you want, and Kyrall handles the rest. Batch generation, design exploration, and iterative refinement, all from a simple prompt.

Until now, macro design modifications had to be done manually Even with design exploration pipelines, everything had to be hard-coded Not anymore… kyrall.com

@ben_sdl @DominiqueCAPaul Very nice event! We definitely need more events like this in Munich

The Demo Night Munich was incredible! The talent density in this one room was genuinely insane. What brought all of us together, was that everyone at the Demo Night is passionate about something hard, building it either on the side or having gone all in. Leaked demo👀

@theworldlabs Was the main scene still designed manually in Blender? What about parametric batch generation of scenes and assets? This parametric scene, for example, was generated automatically from a text description using our tool in just two minutes.

Lay out a scene in any 3D tool, then bring your world to life inside Marble! Go from basic layout → fully realized world in minutes 🌎

Most companies working on improving hardware design are just trying to build a faster horse. They are missing the point

@sammartin @nunzi46 Exactly! People assume that if AI designs something it won’t undergo simulations and testing

@nunzi46 If it flies for 1000 test hours without issues, what’s the difference?

Would you fly in a plane that was designed using Text-to-CAD?

@ben_sdl I will add it to the list of options

@osaAtwi Wispr to CAD?

“Text-to-CAD” is a bad name It massively undersells what’s actually possible. Most criticism is a Strawman; attacking a toy version of the idea that no serious system is trying to build.

@amypretzel This is why CAD itself is not engineering. Much of the work still consists of repetitive manual tasks that should ultimately be automated

hot take: CAD tutorials teach you the software but not how to think I watched hundreds of hours of SolidWorks tutorials in college. none of them taught me that real manufacturing has constraints that will ruin your beautiful model. learned that the hard way when my first injection mold came back wrong

@emm0sh I think people dont know what Text-to-CAD is. Most criticism is a strawman. But I have to say the name "Text-to-CAD" needs to be changed

text to cad, design an F1 front wing. make no mistakes

NO CAD Knowledge NO Manual Work Design Exploration in Mechanical Design is a big topic but also a complicated one Much of the work is done manually Those three examples were generated in less than 1 minute! This means Engineers can now design complex assemblies in minutes

Demos are easy... Show me your assembly Generated with Kyrall in less than 2 minutes! A wing is not only a 3D rendering of the outer shell. It is the spars, ribs, connectors to fuselage, cable harness and much more

Flashy demos are easy. Useful systems are hard. Generating a random 3D model from text with Claude Code or Codex is trivial. Generating a properly sized glider aircraft that could actually fly? That’s what Kyrall is built for.

Kyrall can now generate any 4 Digit NACA profile

There's a belief deeply embedded in engineering culture: the more "Manual" control you have over a 3D model; every radius, every edge, every connection point; the better the outcome. More control equals better design. That is the wrong approach! kyrall.ghost.io/you-dont-need-…

A propeller test bench built around a generic forces and moments measuring unit. Aluminium extrusions form the frame, joined by a central bracket, and stabilized on a table with 3D printed 90° corner brackets.

No CAD experience. No manual modeling. Every single part generated by AI using Kyrall. CAD was just the viewer