Personal Project — AI Infrastructure
Home AI Inference Server
What it is
A self-hosted inference stack running on a home server (AMD Ryzen 7950X, 128GB RAM, AMD 7900 XTX) on Arch Linux — built to explore local LLM hosting without cloud dependency.
Stack
- ROCm for AMD GPU acceleration + Ollama for model serving
- Docker Compose for service orchestration
- n8n for workflow automation on top of the models
- Tailscale for secure remote access
- Grafana for monitoring
- Qdrant for vector storage
- Piper for local TTS
The interesting part
- Explored multi-backend llama.cpp limitations when mixing AMD and NVIDIA hardware — worth a short writeup, since GPU-vendor interoperability in local LLM setups is a real pain point people search for
- Diagnosed and recovered from an Arch Linux boot failure (stale fstab entry) via live USB chroot — a good "I can operate at the systems level, not just the application level" anecdote
This reflects a broader pattern: building interconnected systems rather than isolated scripts — a thread that runs through this project, the Gressingham automation work, and the NorwichVR infrastructure.