A/V engineer in Oklahoma. I build the software I want to see in the world, with AI in the loop. I plan every build, choose the stack, and run it on servers I manage, all inside a development system I designed and built.
Tap play for musicNow playingPausedMusic is not available hereJezero MorningButterscotch AfternoonBlue SunsetStarlit RowsColony SquareFestivalGlimmer Caves
Recorded from the running game; the controls were scripted. The video has no sound. The music is the game’s own, made in code and played live in this page. Tap play to start it.
01Home-theater app
Marquee
Screen recording comingThe app icon is real. The recording is not ready yet.
The real Linux app (C++23, Qt 6/QML, libmpv) browsing a test library: invented titles, test-pattern clips. Driven from the keyboard, the way a remote drives it. 38 seconds, no sound.
Open-source (GPL-3.0) home-theater app for a private local movie and TV library. Remote-first, ten-foot interface.
Role
Planned it and chose the stack, tested and refined it. Shipped it.
Stack
v3: C++23, Qt 6/QML, libmpv, SQLite
Ships
GitHub Actions with license, signature and privacy checks
2,000+commits since March 2026, across the first site and its rebuild
Infrastructure & hardware
I run my own small infrastructure. I build out of old hardware.
What I run
30+repositories on my own Git server, each mirrored to GitHub automatically
A cloud server with a self-hosted Git server (Forgejo)
A TLS reverse proxy (Caddy) that serves the public sites
A private network (Tailscale) that reaches every machine from anywhere
Ad and tracker blocking for the whole home network (AdGuard Home DNS)
A messaging relay service
Nightly backups to a separate disk
Second Chance at life
Old hardware, and Linux on all of it.
2011MacBook ProA broken screen and no software upgrade path. It is now a headless Linux server that runs the home’s DNS filtering. I installed it and configured it over a cable from another machine.
2011iMacMy daily Linux desktop and main build host.
2015MacBook AirA second Linux build and test machine and remote work station.
For Fun: iPhone 12 research
4repeatable ways to crash the boot ROM’s USB recovery from a PC, so far
Goal: Linux on my own old iPhone 12 (Apple A14 chip). Started Oct 2026.
Public jailbreak exploits do not cover that chip, so this is my own research.
I dumped and disassembled the phone’s read-only boot ROM and audited its USB recovery mode (DFU).
A small Python tool tests each suspect sequence on the real phone and logs every USB transfer.
Found: the safe window of requests that do not crash it.
Rules: never boot a foreign image, never write to flash. A crash is data, not damage.
Next: a Raspberry Pi Pico 2 raw-USB test rig, firmware written, that refuses the four known crash sequences by design.
~/How-I-build
This is my own system for software development and AI orchestration. I started working with AI in January 2026, on one MacBook Air, and since then I have spent thousands of hours designing, building and refining it, on my own time and my own money. I plan each build in detail and decide the architecture and the stack. AI coding agents carry out the plan inside the system.
How it all connects
My whole system, top to bottom. As of now.
I decideMePhone and laptop
Plans each build
Chooses the stack
Tests the product
Refines designs, owns releases
Private network (Tailscale): reaches every machine, from anywhere
Home machines
iMacBuild host
MacBook AirBuild host
MacBook ProDNS server
Cloud server
Git serverForgejo. Every repo.
Web proxyCaddy. Serves the public sites.
GitHub
MirrorCopied automatically.
I decidePlanThe outcome, the architecture, the stack, and how we know it is done.
AI coding agentsBuild to plan, in sandboxes. No push rights.
Separate AI reviewerRead-only. Approves or sends it back.
CIMust pass.
I decideReleaseHealth check, rollback, I check it live.
Live sitesOn servers I run.
Black: me, the human in the loopOutline: agents work to my plan
Swipe the steps. Tap one.
01Plan the buildI decide
02Cut it upAgent
03Build in a sandboxAgents and me
04Test and commitAgent
05Independent reviewAgent, then me
06Test and RefineI decide
07ShipI decide
I plan each build in detail: the outcome, the architecture, the stack, and how we will know it is done.
A coordinating agent cuts the plan into small tasks and writes each agent a short task packet: what to do, how to check it.
Each task goes to a fresh AI coding agent in a sandbox: its own Linux user and clone, no private files, no push rights.
The agent runs the tests and commits each change with a changelog entry. Bugs it does not fix go in the repo’s issues list.
A separate, read-only AI reviewer reads the commits. Fixes are new commits by a fresh agent, and review repeats until it approves. When needed I read the code, though not every line.
I check the app looks, feels, and behaves as intended. Or when a design choice is open, agents build two to four throwaway mockups and I pick one and refine it. This site was made this way.
Then it goes out: push, CI must pass, and deploy to servers I run by the repo’s safe path (health check, rollback).
The harness
Harness engineering (Birgitta Böckeler): guides steer an agent before it acts, sensors check after. I built this harness myself and keep refining it. It is model-agnostic on purpose: any capable coding model can fill the agent and reviewer roles.
Guides
Before the agent acts
One rule file every agent reads
A short task packet for every task
A sandbox: own Linux user, no private files, no push rights
A rule is added only after a real mistake; stale rules go
Sensors
After the agent acts
Tests on every run
CI must pass
Pre-push privacy guard
Independent, read-only AI review
Health-checked deploys with rollback
5,000+commits across >30 repositories since March 2026, plus an estimated 6,300+ sandboxed agent runs
1,000sof hours of work building and refining the workflows and apps
Everychange is read by a separate AI reviewer and tested by hand before it ships.
A real failure
FAILRun 1
FAILRun 2
FAILRun 3
SHIPRun 4 passed
The Marquee release build failed three times on GitHub's Mac runners; each failure was read, fixed by a fresh agent and reviewed. Run 4 passed, I verified and validated app behavior and shipped.
/About
I keep complex systems running. Now I build them.
10+years of live sound, lighting, video and music. A professional musician and A/V technician at churches in Texas and Oklahoma since 2011.
1technician for an FAA-approved flight simulator: diagnostics, maintenance, repair, vendor work, FAA configuration documents.
30+repositories on my own Git server, each mirrored to GitHub automatically, with 5000+ commits.
Audio Visual Specialist since Mar 2025, at a federal aviation training site in Oklahoma. Active Public Trust clearance.
Screen recording coming
The app icon is real. The recording is not ready yet.
