I build my entire computing environment from scratch and document it as I go. When something annoys me I fix it. When I figure something out I publish it.
Most of what's here falls into a few buckets:
- IREE/ML research — taking ML service stacks apart, replacing ONNX runtime sessions with compiled VMFB artifacts, and documenting what breaks in the process
- LLM internals — abliteration, disinhibition direction geometry, LoRA fine-tuning experiments
- Homelab infrastructure — AMD/Linux, ROCm, serverless GPU, everything glued together with Python and shell scripts
- Fixes and forks — personal builds of tools I actually use, patched for my setup
Nothing here is maintained. No roadmap, no issue responses, no guarantees. If it's useful to you, great. The documentation is the artifact.
Based in Germany. Self-taught. Works in retail.