Infrastructure / 2025
Homelab Infrastructure
A Raspberry Pi cluster running real services instead of being treated like a toy setup.
Homelab Infrastructure
Design / code / deployment
Homelab Infrastructure
Design / code / deployment
Homelab Infrastructure
Design / code / deployment
Homelab Infrastructure
Design / code / deployment
Homelab Infrastructure
Design / code / deployment
Homelab Infrastructure
Design / code / deployment
Hardware
Raspberry Pi cluster
Services
10+ self-hosted apps
Goal
Cloud independence
Overview
A self-hosted production stack running on Raspberry Pi hardware with Kubernetes orchestration, Proxmox virtualization, and Cloudflare Tunnel networking. It hosts over ten services replacing common SaaS dependencies — including Dokploy for app deployment, Jellyfin for media, MinIO for object storage, Immich for photo management, and Rybbit for analytics — with full operational ownership.
The problem
SaaS subscriptions accumulate recurring cost, reduce privacy control, and create dependency on external uptime and data handling decisions. Running real services on owned hardware forces a level of operational discipline — monitoring, backup strategy, failover planning — that cloud abstractions typically hide.
What I built
The homelab runs Kubernetes on a Raspberry Pi cluster for service orchestration, Proxmox VE for VM isolation and management, Tailscale and NextDNS for secure remote access, and Cloudflare Tunnels for safe public exposure without direct port forwarding. Each hosted service replaces a SaaS equivalent with something under direct operational control.
My role
Complete ownership across hardware procurement and physical setup, Kubernetes cluster configuration, Proxmox virtualization setup, Cloudflare Tunnel and Tailscale networking, service deployment and configuration, monitoring setup, and ongoing operational maintenance.
Biggest challenge
Running production-like services on ARM hardware means tighter memory budgets and more sensitivity to thermal and power constraints. Consumer-grade failure modes — SD card wear, power fluctuation — require backup and recovery planning that enterprise hardware handles automatically. Keeping over ten services stable and observable without dedicated ops time required disciplined automation.
Next project
Fine-Tuned Large Language Model