Project - 2024
InteriorLab.
3D design,
powered by AI.
A terminal app that lets architects, designers, and 3D creators describe what they want in plain English and watch it render in real time - powered by GPT-4, OpenSCAD, and a custom VTK visualization engine.
How It Works
Prompt -> 3D output.
InteriorLab bridges natural language and 3D geometry through a fully automated pipeline. A user types a design prompt, GPT-4 translates it into SCAD code, that code is compiled into STL geometry, and the result is rendered live inside our custom VTK-powered viewer - all within the same app.
The back-and-forth between the AI model and the rendering engine is seamless - you can refine your design conversationally, and the 3D view updates autonomously in response.
Key Features
What InteriorLab does
GPT-4 converts plain English design prompts directly into OpenSCAD code - no manual geometry scripting required.
SCAD designs are automatically compiled into STL format and piped into the visualization layer without any manual steps.
We built our own 3D visualization software using Python's VTK library - enabling real-time rendering inside the app itself.
Users can iterate on their design through follow-up prompts. The AI updates the geometry and the view re-renders automatically.
Challenges
What we had to figure out
Full automation. Making the entire pipeline autonomous was novel - nothing like it existed, which meant we had no reference point for how to wire AI into 3D generation at this level.
SCAD to STL conversion. Getting reliable, error-free geometry out of GPT-generated SCAD code required significant iteration - the model's output needed heavy validation before it could compile.
VTK integration. Embedding the VTK rendering engine into the Tkinter app and getting it to update live on each design change was one of the harder technical milestones of the build.
Impact
Who this is built for
There are over 3.2 million architects worldwide, plus a broader audience of 3D designers, interior specialists, and 3D printing enthusiasts who currently rely on complex, manual tooling to visualize their ideas.
InteriorLab makes 3D design accessible through conversation - removing the barrier of learning CAD software and letting creators focus on the idea rather than the tool.
What's Next
Where it goes from here
The next step is moving InteriorLab out of the terminal and into the browser - making it accessible to anyone without a local setup. A web-based version would open the tool up to a much wider audience and set the foundation for taking this further post-hackathon.