Become a sponsor to Leo Camus
I'm Leo — a self-taught AI infrastructure and full-stack engineer from Paris. I build systems at the edge of what's technically possible, entirely self-hosted: multi-GPU LLM inference on AMD ROCm, open-source intelligence tooling, active-defense security, and independent AI research — every layer built from scratch, from silicon to inference.
Most of my work is open source and free to use:
- Orodruin — a self-hostable, open-source alternative to Palantir Gotham, live at orodruin.dev
- Excalibur — active-defense deception middleware for APIs, with a playable live demo
- ROCm inference stacks — FLUX and multi-GPU diffusers at parity with NVIDIA on AMD hardware
- CAMUS Theory — independent, published research on temporal cognition in LLMs
Sponsoring directly funds this work: keeping the live demos online, covering compute and hardware, and freeing up time to keep shipping open-source tools and research instead of locking them behind a paywall.
Every contribution — at any tier — genuinely helps, and keeps this independent. Thank you for supporting open, self-hosted, no-compromise engineering.
Featured work
-
pytorch/ao
PyTorch native quantization for training and inference
Python 2,964 -
Dev-next-gen/orodruin
Orodruin — open-source intelligence platform. A self-hostable Palantir Gotham alternative: dozens of live global data sources on a 2D/3D map + actor graph, with an AI analyst that drives the UI.
JavaScript 18 -
Dev-next-gen/excalibur
Active-defense deception middleware for SaaS APIs: serves booby-trapped fake data to attackers, logs and geolocates them. Rust SDK + ML control plane (PASS/CANARY/DECEIVE) + canary tracer + SOC das…
Python 2 -
Dev-next-gen/flux-amd-rocm
FLUX.1-dev on AMD Radeon consumer GPUs — fast, low-VRAM, and shippable. Backport patches + benchmarks for torchao + diffusers group_offload on ROCm.
Python 2