torvalds gitline

I recently launched Gitline Gifts — a service that transforms developers’ GitHub contribution graphs into 3D sculptures.

The Inspiration

Inspired by my friend Nathaniel’s experience with 3D printing, I wanted to build a product that was interesting but also highly individual - something that couldn’t be done using traditional manufacturing means.

Learnings

2D->3D, aesthetics, STLs

AI agents continue to surprise with how good they are at generating plausible outputs. In the case of Gitline, we were able to zero-shot transform github contribution histories into STL files for 3D printing and it mostly just worked. However making any changes beyond that is incredible painful - tasks such as adding simple components, moving objects around, etc - the LLM tries very convincingly but ultimately any adjustment required falling back to the tried and true ‘move code around, test, validate manually’.

Moving across agents

I started out with Repl.it because of their all-pieces-provided approach – however when we get deeper into building features I moved away because I wanted a more fully fledged editor environment. Exporting the project from repl.it and spawning a claude agent on top of it was almost okay but it took more than a little effort to get Claude to wrap its head around the project. Probably better to start with one tool and use it through the end for now.

Mocks

Agents love generating mock data and demo modes – I don’t know where this comes from but multiple times I had to explicitly instruct them to “build the feature right” rather than work around real issues by just falling back to mock data.

Conclusions

Overall this was a fun project that gave me a good window into the state of AI agents - they continue to be so impressive in so many ways which makes their failings all the more surprising & frustrating!

Check out Gitline Gifts to transform your own GitHub contribution graph into a desk-worthy conversation piece.