Senior Software Engineer // C++ · Python · Bash · Linux · DevOps
Seven years of real-time C++ and Python in defense and aerospace. I built the back end of a real-time radio transceiver, wired a neural network into a multi-camera thermal-imaging array, worked on GPU-accelerated radar detection, automated a Kubernetes-based DevOps platform across AWS, Azure and OpenStack, and was the sole developer of a satellite-to-ground link simulator.
At home, my Linux desktop is a lab. AI coding agents that I direct build and run my tools there: an alloy simulator, an automation hub, sandboxed agent machines, and patches for the KDE desktop I use every day. I set the goals, pick what to chase, and review what comes back.
A physics pipeline that hunts for new metal alloys on a desktop PC. A machine-learned interatomic potential (MACE) relaxes atomistic cells and works out stiffness and phase stability for thousands of candidates, then ranks them. Validated against 1,214 known alloys: the right crystal lattice for 99% of those with one right answer, and a 1% median density error.
The control room of my desktop: a Bash loop and a PySide6 window that run queued jobs and one-shot hooks, pass work to three AI assistants through plain-text pipes, keep a process registry and a cgroup CPU governor, and sync with my second PC through a private Fossil repository.
Headless AI agents, each in its own rootless Podman container on a dedicated NVMe drive. AutoShell feeds them a prompt queue; every run gets its own log and results, the model is pinned, and updates go in under a lock. The first core's first job: redesign the file structure of the hub it runs in.
Two bugs from my own desktop, traced to their source and patched. In KWin's Wayland session, a panel on a smaller monitor clipped the main monitor's work area, so Steam's notifications sat 360 px too high. In bluez-qt, sending several files over Bluetooth crashed the sender (KDE bug 525109).
A 3D mech-combat game in Godot 4, made with other developers: mechs assembled from a data-driven parts catalog, player and AI controllers, and a headless test harness that checks the game without a screen.
Tools for my Kick channel, The TANgent. It reads chat aloud in an AI voice with a censor in front of it: built-in rules, my own block and allow lists, and a language model that judges each message. Plus a one-click Go Live that starts streaming and recording in OBS.
BUILD_NOTE // The simulator, AutoShell, the Imagination Cores, the KDE patches and the Stream Assistant were written by AI coding agents (Claude, Gemini, Grok) that I direct: I set the goals, choose what to pursue and review the results. Project Crash is AI-assisted.
Email is the quickest way to reach me. The code is on GitHub, and my work history is on LinkedIn.