[AI] Claude Sonnet for writing a Loading Function for X3D.py
Bergstrom, Aaron
aaron.bergstrom at und.edu
Sun Mar 22 15:09:26 PDT 2026
All,
I don't like writing loaders. And I haven't had the best luck using an LLM to write code for me.
However, after attending NVIDIA GTC this past week, I realized that I would be doing my student software engineer employees a disservice if I didn't have them incorporate the use of LLMs into their code projects for USD2X3D and RawKee for Blender.
We are trying to teach them industry standard coding practices, and leaving out the use of LLMs in their coding practices seems to me to be shortchanging my students on the one set of skills that will help them get jobs after graduation.
I had been using the free versions of stuff with the work I had tried to do in the past... so you know... they were several different iterations of LLMs behind the current versions.... So my prior results probably aren't an indication of what's actually possible.
As it turns out, the DREAM Lab has access to a bunch of tokens of various LLMs through our GitHub Enterprise account, which I was not aware.
So this weekend, I had Claude Code - specifically Claude Sonnet 4.6 evaluate my custom RKx3d.py file, and my RKSceneTraversal.py file (which contains my "x3d2disk()" function), and had it write a "disk2x3d()" function for loading X3D XML, CLASSIC, and JSON files into an x3d.py scenegraph.
While I haven't yet had a chance to test it, just a quick review of the python code that Claude Sonnet produced suggests that it did a pretty good job, and I think it got it mostly correct, with the possible exception of Script/Shader <field> tag and CDATA values.
I've got a fair amount of HPC work backlog to grind through since I've basically been traveling for past two weeks, but hopefully I'll be able to test and integrate this new code in the next couple of weeks, and then made significantly accelerate my exporter work and the work of my students with the use of Claude Sonnet going forward.
Aaron
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://web3d.org/pipermail/ai_web3d.org/attachments/20260322/82995634/attachment.html>
More information about the AI
mailing list