I build practical local AI, robotics, and developer tools.
I like projects that can be reproduced on ordinary hardware, inspected end-to-end, and actually used โ not just demos hidden behind an API.
-
microduck-vision-soccer
Autonomous robot soccer in MuJoCo for Pollen Robotics Microduck: head-camera vision, visual servoing, physical biped kicks, and a monocular goalkeeper. -
physical-ai-playground
Experiments with Gemini Robotics ER 2, SO101, MuJoCo, and practical embodied-AI workflows.
-
private-knowledge-base
Local-first RAG with real source citations, BGE-M3 reranking, Qdrant, Ollama, a React UI, reproducible benchmarks, and an 8GB-GPU-friendly setup. -
qwen-image-2.1-8gb
Run Qwen-Image-2.1 on an 8GB NVIDIA GPU with ComfyUI + GGUF, ready-to-use workflows, and a VRAM doctor.
- imdb-monitor
A self-hosted movie / TV tracker with IMDB/TMDB integration, notifications, web UI, and Docker deployment.
Current focus: local models, physical AI, robotics simulation, and tools that make new AI capabilities usable on consumer hardware.
If something here is useful, issues and PRs are welcome.