All case studies
02 — Own project · the homelab as a lab
A private cloud for AI and automation
Own on-premise infrastructure: virtualisation, a TLS reverse proxy, external access without open ports, n8n, a vector database, monitoring and backups.
Problem
Companies and institutions want to use AI without sending their data to the cloud, but don't know how to build and run such an environment. I needed a place where agents, automations and vector databases actually run — not just on a slide.
What I built
- Virtualisation cluster — three Proxmox VE nodes, 34 VMs and containers (24 running), nightly backups to Proxmox Backup Server.
- Traffic and access — a Traefik reverse proxy with wildcard certificates issued through a DNS challenge. About 39 services run behind Traefik. Internal services are reachable only from the local network or over VPN (Tailscale); selected services are published through Cloudflare Tunnel with rate limits, without opening router ports. Firewall: OPNsense.
- Services for AI and automation — n8n, Qdrant, Redis, MinIO, GitLab, Ansible (Semaphore).
- Observability and secrets — monitoring with Prometheus, Grafana and Loki; secrets in Vaultwarden, which my agents use as well.
- A guide: "Deploying LLMs on an internal network" — inference engines (vLLM, SGLang, Ollama), the UI and RAG layer, Polish language models (PLLuM, Bielik).
Outcome
- One environment for agents, automations and vector databases — with backups and monitoring.
- My other projects run on this infrastructure, including the mail microservice that delivers this site's contact form.
- Hands-on experience I bring to on-premise deployments and training.
Facing a similar problem?
Describe your process — I'll tell you whether and how it can be solved.