Before

In my university years, I was somehow involved in 3D printing, mostly through my friends and scientific association. That said, it has never caught on as a hobby; CAD softwares were quirky, printers were expensive, and were 'problematic to live with' types of device.

Now

I've recently purchased a BambuLab p1s - enclosed type of fused deposition modeling printer, capable of printing multiple materials. Why Bambu? Why not Prusa? Being honest, the enclosed construction was deal breaker for me - minimising potentially dangerous fumes from printing was priority; coupled with https://www.printables.com/model/272525-bentobox-v20-carbon-filter-for-bambu-lab-x1c-enclo and frequent room airing, I am less concern about my respiratory health. Also, BambuLabs are known for its low maintenance requirements. And design kind of match my home-office vibe.

Problem

Printing is easy, but creating your own designs? That's totally different story. I've tried several free tiers of CAD software, and fusion360 and Shapr3d are the best so far. Especially the second one, more welcoming learning path, using ipad with pen was much appreciated. But nevertheless, both softwares are expensive, and, for free tiers, have limitations. Shapr3d does not support round objects to be nicely exported to STL format, which is a bummer. And fusion is industry, not a hobbyist level (I admit some features, like generative design are extremely cool). My first creation, the airtag holder fitting under bike bottle cage, took me 10h trans Atlantic flight; from 0 experience to printable object with some design flaws.

Solution. Sort of.

Is it possible to write down physical object idea on your computer and within 1 hour get yourself an exact object on you desk? I am delighted to inform you that: it is fully possible, but it might be frustrating in the process. Especially if you are doing it without checking the design.

As part of the task, I needed code-based CAD and I found one, probably the most popular out there - Openscad https://openscad.org/. It’s well known solution and definitely the learning curve is more gentle, but also not trivial, especially for more complex shapes.

Combining openscad with most popular publicly available LLM, ChatGPT 4o, is quite simple. Without spinning any docker, feeding any data or using any knowledge for both subjects. Just free tier account, verbalised idea and lots of hope.

using openscad create me a airtag holder, that have M5 screws mounting holes with 64mm between, and eliptic shape, with one side blind

After 5 iterations, of me saying 'that's basically a sphere, not an airtag holder' and 'that won't fit an airtag', I got expected shape

After adding some features and more precise descriptions, forming kind of a list of features that needs to be incorporated to the final design, the LLM came up with following solution. Also I've hit pay-wall for free users.

Big remark: the objective of this activity was not to make any manual changes, as soon as shape was already matching my expectations - I've exported STL file.

Printing time

The process of creating artificially made cad project, to create a real object, all within 1 hour, gives me chills and futuristic feel. I played 'Robot rock' by Daft Punk, and hit ‘printing’. Forty minutes later, AirTag holder was done.

Model review

As you can observe, even after few iterations, design is still flawed; namely a thickness of the walls included as part of inside compartment, not as outer body. That resulted in slightly bigger space for AirTag, which did not hold it securely and made a rattling noises. Also, whole design is not what you would call economically optimised (it's big), and not necessarily matching description (it's stadium shape, not elliptical). Despite these minor, easy to fix issues, the design was something that I had on my mind before starting.

Here is a bonus, my design compared to LLM design. I've created lid for maximum weatherproofness, as well as some regions designed as empty space to reduce amount of material required.

Outro

I admit, my own CAD design is nice and functional, but LLM design isn't far off, especially with some minor adjustments. For normal, day-to-day functional prints, that might be pretty useful technique.