I'm Cameron — IT Systems Analyst at Redwire Defense & Space on the Central Coast of California. I manage the firewalls, networks, servers, and ZScaler infrastructure that keeps engineers doing their jobs without noticing the stack underneath. On the side I consult for local businesses — helping places like Adelaida Winery keep their networks and servers running.
Outside of work I'm in the homelab. What started as gaming-servers.net has grown into a K3s cluster running a dozen services — the biggest being ipinfo.app, an IP intelligence suite that has handled over three billion requests across ASN lookup, proxy detection, and geo tools. I also build and maintain real estate platforms, music preservation archives, cosmetologist portfolios, and whatever else sounds interesting.
When I'm away from the keyboard I'm out in nature — hiking trails, taking in views, and finding the kind of quiet that only exists outside. I hunt down craft beers and wines across the Central Coast, and burn vacation club points on trips to Hawaii and Bend, Oregon — a purchase I'd describe as equally stupid and awesome.
My hypervisor sat at 40% iowait for days and survived every reboot. The SSDs were healthy, the RAID controller was healthy, memory was idle — and the real cause was a kernel I'd pinned during a major upgrade and never moved off. A walk from iostat through PSI to proxmox-boot-tool, and why a 2026 QEMU on a 2024 kernel manufactures phantom iowait.
Caddy 2.11.4 landed on June 3rd — a security-heavy patch release with four hardening fixes and a pile of TLS and reverse-proxy bugfixes. Here's what changed, the one breaking caveat worth reading before you upgrade, and how I rebuild it as 47 ready-to-pull DNS-provider images for DNS-01 ACME.
During COVID I bought three XPS 17s to game over Parsec. They were flawed — an iGPU memory leak that ate 128 GB of swap, a battery that drained while plugged in — and they're what pushed me to a Mac. Years later one of them, sitting idle on my desk, became prb-bld-01: a dedicated CI build node that cut a 40-minute build to 10. The throttle hunt, the battery cap, two runners on one box, and saving a consumer SSD from CI.
Chrome quietly rolled out Gemini Nano — a ~4 GB on-device LLM that downloads the first time any page calls the new `LanguageModel` API. Here's what it actually is, what it can't do, and what happened when I spent an evening building a search-and-fetch agent loop on top of a 3-billion-parameter quantized model.
Finally putting two Tesla P4s to work — passing them through Proxmox to a small VM, running ollama, and exposing it to my apps via Cloudflare Tunnel with Bearer-token auth. What tiny models can do on hardware that's been collecting dust for years.