I’ve been using Large Language Models (LLMs) for a while now. Whether it’s writing complex Go services, drafting documentation, summarizing long reports, or even designing AI agent sandboxes, LLMs have become my go-to co-pilot. But recently, I decided to push the boundaries a bit further.
What if I could use an LLM not just to write code that runs on a CPU, but to generate physical objects?
As an automobile enthusiast and an avid collector of 1:64 scale diecast cars (yes, my desk is covered in Hot Wheels and Tomica!), I had the perfect challenge: designing a Modular Diecast Display Case. And I wanted to do it entirely through a parametric, code-first CAD approach powered by AI. 🛠️
🎨 The Vision: A Perfect Home for 1:64 Cars Link to heading
If you’re into diecast collecting, you know the struggle. You want to display your favorite cars, but standard cases are either too bulky, too expensive, or don’t stack well. I wanted something specific:
- Modular & Stackable: So the collection can grow.
- Minimalist: To keep the focus on the car.
- 3D Printer Friendly: Optimized for zero-support FDM printing.
- Parametric: So I can tweak dimensions easily if I ever switch to 1:43 scale!
⚙️ The Technical Stack: Code-First CAD Link to heading
Instead of traditional “point-and-click” CAD software, I used build123d, a Python-based parametric geometry library. Why? Because LLMs are incredible at writing Python.
To make the workflow even smoother, I utilized a specialized “text-to-cad”
agentic skill. This allowed me to describe the geometry in natural language,
have the agent generate the build123d Python code, and then iterate on the
design programmatically.
📐 Designing the Geometry Link to heading
The result of this AI-assisted collaboration is the diecast-modular-display. Here’s how the technical details broke down:
1. The Core Shell Link to heading
The case is a hollow box shell with dimensions of 84mm (W) × 36mm (D) × 39mm (H). I went with a 2mm wall thickness to ensure it’s sturdy enough for stacking while remaining lightweight. The front face is completely open, providing a perfect side-profile view of the car.
2. Built-in Car Stabilization Link to heading
Nothing is more annoying than a car rolling around inside its display. I added a 1.5mm deep sunken wheel tray (72mm × 26mm) on the inner floor. It’s a small detail that makes a huge difference!
3. Support-Free Optimization Link to heading
To make printing a breeze, the model is designed to be printed lying flat on its back wall. I added stylized diamond cutouts on the back and sides. By using 45-degree angles for these cutouts, the printer can bridge the gaps without needing any messy support structures. Clean prints, every time! ✨
4. The Stacking Mechanism Link to heading
To make it truly modular, I implemented a pin-and-pocket system:
- Top: Two Ø5.0mm × 2.0mm locking pins.
- Bottom: Two Ø5.4mm × 2.2mm receiving pockets.
That 0.4mm diameter tolerance (0.2mm per side) is the “magic number” for most FDM printers to get a snug, satisfying snap-fit.
🤖 The Agentic Workflow Link to heading
What made this project special wasn’t just the final STL file, but the process. Using the earthtojake/text-to-cad skill, I could verify the geometry autonomously. My “Agent” could:
- Generate the
build123dPython script. - Compile it into STEP and STL formats.
- Inspect the topology to ensure the faces and planes were correct.
- Snapshot the model from an isometric view to show me the progress.
It felt less like “drawing” and more like “directing” a master craftsman.
Final Result: A Sleek, Functional Display Case Link to heading
🚀 Get the Files Link to heading
The entire project is Open Source. Whether you want to print some cases for your own collection or dive into the Python code to see how the parametric geometry is built, you can find everything below:
👉 GitHub: anshulpatel25/diecast-modular-display
👉 3D Files: Diecast Display
🏁 Final Thoughts Link to heading
Moving from “Text to Code” to “Text to CAD” feels like a natural evolution. LLMs aren’t just for chatbots; they are becoming powerful engines for spatial reasoning and engineering. Designing this display case showed me that with the right tools and a bit of enthusiastic prompting, we can bring our digital ideas into the physical world faster than ever.
Now, if you’ll excuse me, I have a few more cars that need a new home. Happy printing! 🏎️💨