- 8
- Projects documented
- 47
- Posts published
- 5
- Deep-dive series
- Since ’23
- Writing in public
Selected work
Featured projects
Systems I built end to end, each one written up.
Project · 2026
Project Aether: A Three-Environment Homelab Platform
A production-style platform on a repurposed laptop and two free-tier clouds: K3s with GitOps, Terraform and Ansible for every environment, a Tailscale zero-trust network, encrypted secrets in Git, and tested backups.
- K3s
- ArgoCD
- Terraform
- and Tailscale, Proxmox
Project · 2026
OpenClaw on Azure — Zero-Trust AI Agent Deployment
Deploying a personal AI assistant on Azure with Zero-Trust networking, Terraform IaC, Managed Identity, Key Vault secrets, and Tailscale mesh VPN.
- Azure
- Terraform
- Tailscale
- and Key Vault
Project · 2026
Cloud Resume Challenge (Azure)
Full-stack serverless portfolio on Azure — Hugo frontend, Python API, Cosmos DB, Front Door CDN, Terraform IaC, and GitHub Actions CI/CD.
- Azure Functions
- Cosmos DB
- and Front Door, Terraform
Writing
Recent writing
Production issues, the architecture choices behind them, and what actually worked.
homelab
Sizing Kubernetes on a 12 GB Laptop: What the Scheduler Sees and What's Really There
Live numbers from a three-node K3s cluster on a 12 GB laptop: requests, limits and real usage per node, the control plane the scheduler can't see, the limits that OOM kills corrected, and the gap I found while writing this (nothing reserved for the OS, no memory eviction at all) with the same-day fix.
homelab
A Free Outside Observer: Running My Homelab's Watchtower on an IPv6-Only GCP VM
My homelab is watched from a free Google Cloud e2-micro with no public IPv4 address. What it does, what going IPv6-only broke (including eleven days of alerts that never left the host), fitting it into 1 GB of RAM, and why the least-trusted machine got the least power.
homelab
Observability Off the Cluster: Prometheus and Loki in the Cloud, One Agent per Node
My homelab's monitoring runs on a free cloud VM, not on the cluster it watches. Why the split, what each node runs, the memory budget that turned out wrong, swapping Promtail for Grafana Alloy without a gap, and the alerts that watch the monitoring itself.