Home Lab — own servers, own stack

Use case

A small private infrastructure that runs all my projects and data without vendor cloud. Authentication is centralized via forward-auth SSO, DNS and ad-blocking via Pi-hole, smart-home and voice input via Home Assistant with Wyoming services, container workloads via Docker on the bigger machines. Everything sits in a home server rack with UPS, switch, and tidy cabling I've grown attached to.

Hardware

• AI Server — AMD Ryzen 5 4500, 31 GB RAM, with two mid-size GPUs (GTX 1060 6 GB + RTX 2060 6 GB). Docker host for most services.
• LLM Server — separate machine with a modern 24 GB consumer GPU for serious LLM inference with quantized open-weight models from the qwen family.
• Raspberry Pi cluster — several Pis in service, including one as reverse proxy with forward-auth and one as CCU bridge (debmatic) for Homematic IP.
• Switch + UPS — small Layer-2 world in the rack, UPS for the critical nodes.

What's done so far

• Authentication: forward-auth SSO as central gate in front of all internal services
• DNS: Pi-hole filters for the whole LAN, plus internal domain resolution
• LLM inference: Ollama on the LLM server, hybrid routing between local models and external providers, bridge layer for agent integration
• Smart home: Home Assistant with Wyoming services and Faster-Whisper (large-v3-turbo) for voice processing
• Observability: Grafana with a custom look + TimescaleDB as central time-series database
• Database layer: multiple PostgreSQL instances per project domain
• Vector store: Qdrant for embeddings and RAG experiments
• Reverse proxy: TLS termination via Let's Encrypt wildcard, routing to service containers on the LAN

In progress

A cleaner cross-machine backup strategy, capacity planning for GPU load when multiple AI projects pull at the same time, a few more Wyoming services, and a refactor of firewall rules. Smart-home, health, legal, and trading stacks all run on top of this.