CS grad (Penn State) and full-time IT Support Specialist keeping several regional campuses running — but the title undersells it. I architect cross-campus asset-reconciliation systems and spend the off-hours co-founding Susquehanna Syntax and building a self-hosted AI stack in my basement.

Open-core by default. Everything ships under one house style and a stubborn -il naming convention. Based in NEPA, trying to turn the Wyoming Valley into a place you don't have to leave to do this work.

role      →  IT Support Specialist  ·  cross-campus systems  (Penn State)
building  →  Susquehanna Syntax  ·  open-core, self-hosted, design-led
homelab   →  AMD MI50 32GB · ROCm/gfx906 · llama.cpp router · ~48–51 t/s
cert      →  Certified Extron Control Professional
elsewhere →  NEPA transit & regional-dev nerd  ·  Trains are Awesome and WILL happen

_{Open-core. One design language across all of it: charcoal base, six pastel accents, Fraunces / DM Sans / IBM Plex Mono, flat and calm.}

• Home inference stack — AMD MI50 32GB (gfx906) on ROCm, rocBLAS rebuilt from source, llama-server running in router mode as a systemd unit with MTP/speculation passthrough. Benchmarking ~48–51 t/s.
• Agentic loops — Claude-as-Architect → local-model-as-Executor overnight coding runs via a homegrown MCP bridge.
• A publishable idea — token-level Mixture-of-Models routing between two GGUFs that share a tokenizer. Current tooling doesn't do it; I'm poking at the gap.
• The boring-but-real stuff — mergerfs + btrfs RAID1, Caddy reverse proxy, Gluetun, Portainer, and a hard-won respect for not running snap Docker near a Postgres WAL.

I write transit proposals for fun. There's a real one — the Trillium Line, a multi-color regional rail concept (RBMN partnership, public-capital estimate and all) aimed at the LCTA and the NEPA MPO.

Also: Minecraft modpack dev (Entropy³, GregTech/KubeJS), and a weakness for well-made tooling.

Stats