Self-hosted hybrid cloud
Multi-AZ Kubernetes on Proxmox, declared in git and rolled out by Ansible, ArgoCD, and GitHub Actions. The same patterns I use at work, sized for a basement.
The workshop behind
the résumé.
Side projects in three groups. The platform under everything. The spots where AI is doing real work, not driving the brand. The apps I ship through the same pipeline that serves this page.
01/Group
Built on bare metal
Infrastructure and platform work. A self-hosted hybrid cloud.
Zero-trust home network
Deny-by-default, allowlist-based access model with VPN, VLAN segmentation, and DMZ firewalls.
Multi-AZ NAS with 3-2-1 backups
TrueNAS and ZFS, with Cloud Sync and RSync handling durability and archival across availability zones.
02/Group
AI as a tool, not a product
Where AI does the actual work: ops, automation, observability. Quietly, in the background.
A local LLM watching the monitoring stack
A local LLM stands up Prometheus and Grafana with Ansible, watches what's actually firing, and tunes thresholds and scrape intervals based on what it sees. Same loop a human SRE runs, just cheaper to run overnight.
AI-enabled home automation
Home Assistant, N8N, and a local LLM running together for private automation that doesn't round-trip to anyone's cloud.
03/Group
Apps I ship from the same lab
Workloads deployed through the same GitOps pipeline that serves this site.
kian.coffee
The page you're reading right now. Angular 21 with standalone components, Tailwind v4, SSR + ISR cache, served out of the same Talos cluster the /lab page describes. If it loaded, the homelab's alive.
Hausparty
A content-discovery app backed by a staged ingestion pipeline. Next.js 16 on the frontend. CloudNativePG for durable state. Dragonfly for rate limits and cache. Ships through the same GitOps pipeline as everything else in the lab.
Pelican + Wings game servers
Pelican control plane on the core cluster. Wings daemons on a dedicated node out on VLAN 80. Friends spin up their own game servers without texting me first.