Web Design with Ai reposted this
Designmatix Companion v1.5.1 is live (now signed) + TTC Web App v1.6 Preview: Here's a behind-the-scenes failure that shaped how we scoped v1.6. Most AI-CAD tools, & including Designmatix's Text-to-CAD (TTC) generate a geometry from primitives. The LLM writes code that calls extrude, cylinder, cut, boolean ops. That works for many shapes. It fails badly for some complex ones. Bevel gears are one of those failures, and so are related transmission components: bearings, sprockets, rack and pinion. Real bevel gears need involute tooth profiles, root fillets, addendum/dedendum scaled to module, and cone-angle calculations. The math is well-established. Ask an LLM to reproduce it from scratch and you get "gear-shaped" output — hollow cylinders if you're lucky, vaguely toothed disks if you're not. The fix isn't a better LLM. The fix is the LLM calling a purpose-built library that already knows the math. TTC v1.6 is scoped to add exactly that: an intermediate layer between the user's prompt, the LLM, and CadQuery, so for gears and similar parts, the model composes calls into libraries built for the job instead of reinventing the geometry. In my own testing this week, that approach took bevel-gear output from unusable to manufacturing-quality involute teeth as demonstrated in screenshots This is a real engineering principle, not an AI one. We don't ask an LLM to write a database from scratch when we have Postgres. Same logic applies to parametric CAD. Today: Designmatix Companion v1.5.1 ships signed. The "Unknown publisher" SmartScreen warning from v1.5.0 is gone. Install now behaves like any other signed and trusted Windows software. Combined with this week's TTC v1.5 backend update is, the .py + .json bundle format that Companion expects — the full pipeline now works end-to-end: TTC web app (v1.5) → bundle → Companion (v1.5.1) Form editor → STEP → Your CAD Tool Try it. Tell me what breaks. Download links in the comments. #MechanicalEngineering #CAD #CadQuery #SolidWorks #ParametricDesign #AIforEngineering