Stunad Engine
A generative CAD system where AI translates design intent into executable geometry graphs.
Try the Engine
Stunad Engine is available as an experimental web tool where users can generate geometry through natural language prompts and live code editing.
The system translates design intent into a structured geometry operations graph which is executed by the kernel in real time.
Note: The experimental server may occasionally be offline, reach me out in case it does not respond.
Overview
Stunad Engine is a generative CAD platform that bridges natural language intent with deterministic geometric kernels. It bypasses traditional GUI menus in favor of a **Language → Geometry Graph → Geometry** pipeline.
The system architecture consists of a high-performance **C++ geometric kernel** (integrated via OpenCascade), coordinated by a Python orchestration layer that executes structured geometry graphs through the C++ kernel.
Core Idea
Traditional CAD tools expose operations through graphical interfaces. Stunad Engine explores an alternative model:
Language → Geometry Graph → Geometry
AI generates a structured operations graph describing how geometry should be constructed. The graph is executed directly by the kernel to produce parametric 3D models.
System Architecture
The geometry operations graph is executed through a dependency DAG within the C++ kernel. This allows deterministic execution and incremental recomputation when parameters change.
- Custom C++ Kernel
- Dependency DAG Executor
- Python Bindings
- AI Geometry Graph Generator
- FastAPI Orchestration
- GLTF Export
- Web-based Viewer
The Hard Problems
Building a generative CAD system isn't just about calling an LLM API-it's about ensuring geometric reliability in an unpredictable execution environment.
- Kernel Hardening: Implementing validation layers in C++ to prevent geometric failures from malformed code.
- Custom Bindings: Using pybind11 to map complex OpenCascade BRep operations to a lightweight, AI-friendly Python API.
- Context Awareness: Engineering prompts that understand spatial constraints and axis-aware design patterns.
- Real-time Synchronization: Bridging the gap between server-side Python execution and client-side GLTF visualization.
Interactive Workflow
The system allows a user to generate geometry through conversation, inspect the generated operations graph, and edit it directly.
Parameters extracted from the operations graph automatically become sliders in the interface, allowing real-time parametric exploration.
Current Status
Stunad Engine is an ongoing research project. The focus is currently on system architecture, generative workflows, and exploring how AI can augment technical design tools.
Why This Matters
The project explores a future where design tools are not static interfaces but programmable systems that collaborate with designers.
By combining AI reasoning with parametric geometry kernels, Stunad Engine aims to redefine how spatial tools are built and used.