Experiment: an ontology-guided design synthesis layer built on text-to-cad

[Marcus-QL-Zhu]

Hi Jake,

First, thank you for building text-to-cad. I have been using it as the CAD and artifact foundation for an experiment in ontology-guided mechanical design.

I built three generative mechanical-watch patterns on top of text-to-cad. The system combines:

• explicit component roles and assembly semantics;
• topology-specific design synthesis;
• coupled geometry and motion constraints;
• build123d assembly generation;
• semantic, material, motion, and validation sidecars;
• failed-seed rejection and isolated retry.

Project:
https://github.com/Marcus-QL-Zhu/Ontology-based-Watch-Generator

Architecture:
https://github.com/Marcus-QL-Zhu/Ontology-based-Watch-Generator/blob/main/docs/architecture/computational-engineering.md

Upstream-integration branch:
https://github.com/Marcus-QL-Zhu/text-to-cad/tree/codex/watch-generator-upstream-integration

The watch generator is a proof of concept rather than a proposed final standard. The larger idea is to define a lightweight, extensible ontology contract for text-to-cad, so domain packages can declare component roles, interfaces, constraints, intended motion, and validation evidence in a consistent way.

Would you be interested in exploring whether a small experimental semantic interface belongs in text-to-cad?

If the direction is useful, I would prefer to start with a small RFC and a minimal schema/API rather than proposing a large merge. The watch generator can remain an external reference implementation while we and the community refine the generic layer and contribute additional domain patterns and validators.

I would be very interested in your feedback on where this layer should live and what the smallest useful upstream contract might be.