Homelab Infrastructure Platform

What this proves: Infrastructure — two servers, 40+ containers, VLANs, and Cloudflare tunnels, all of it having to work at 3am.

Overview

This is the foundation everything else runs on. Two servers, Unraid OS, 40+ containers, and a Unifi network stack that keeps it all connected and secure.

The AI inference cluster (dual RTX 3090s) lives on one server. The other runs the bulk of the container workloads: media, document processing, photo storage, home automation, and the self-hosted services that power the blog and analytics.

Server 1 — AI / Compute

• AMD Ryzen 9 5950X, 128GB RAM
• 2x NVIDIA RTX 3090 (24GB VRAM each)
• Runs: vLLM inference, agentic AI workloads, model training

Server 2 — Production / NAS

• AMD EPYC 4345P, 64GB ECC RAM
• RTX 3060, 8x16TB Unraid array
• Runs: 40+ Docker containers, Docker registry, media, storage

Network

• Unifi Dream Machine Pro (or equivalent gateway)
• Unifi access points throughout the house
• VLANs for IoT, guest, and main networks
• Firewall rules enforcing network segmentation
• Cloudflare Zero Trust tunnels for secure external access

Container Services

• AI inference (vLLM, Ollama)
• Media management (Plex/Jellyfin)
• Document processing
• Photo storage and management
• Self-hosted analytics (Plausible)
• Docker registry
• Blog hosting (Hugo + Netlify/GitHub Pages)
• Home automation

Philosophy

Boring infrastructure is reliable infrastructure. Unraid over TrueNAS because it works. Docker over Kubernetes because it’s a homelab, not a startup. Cloudflare tunnels over port forwarding because security matters. The goal isn’t to run the most impressive stack — it’s to run a stack that works, stays up, and gives me full control over my data.