I build things - and learn a lot by building. Lately that's React stuff and systems. Although machine learning sneaked in there. I like to deeply unterstand what I'm building and how the tech behined that works!
- Thaddeus — the new age of git or at least some primitives. Probbably failling spectacular, but I have some ideas, and it's worth a try. OSS check it out! 👍
- Veyra — Veyra is/was my try at game engines. Cause as most, I'm a gamer by heart. But don't have much time for it. Probably OSS it soon, so check it out.
- Dichter & Denker — a GPT trained from scratch on the German classics: custom tokenizer, ~13M-token corpus, two-stage training, fine-tuned into a chat model you can talk to as Goethe, Kant, Schiller & co. · model on 🤗
- Cat vs Dog Classifier — ResNet50 transfer learning (~94% accuracy) with Grad-CAM explanations, plus a beginner's guide that walks through building it end to end. · model on 🤗
Also building, mostly private for now: a C++/Vulkan game engine, reinforcement-learning self-play agents, and some Claude-based tooling.
Right now I'm trying to reimagine source control from the ground up with Thaddeus. I think it's interesting, and only feature knows where it can lean. And got into Effect TS recently, so much relearning & so powerful.