I turn ambiguous problems
into software that ships.

Selected work

Nine applications · ~90,000 lines of first-party code

FIRST LIGHT

A story-driven arcade game with 132 hand-built levels, four narrators, and an original 36-song score. Releasing on Steam September 1, 2026.

ShippingTypeScriptTauri / Rust

Formal winnability prover · 84 tests · bot-derived difficulty curve

OpenEar

Real-time captioning and translation into ~200 languages, running entirely on one local machine. No cloud, no app, no audio leaving the building.

LivePythonCTranslate2

Model chosen by a 124-minute, three-model evaluation

OpenOrder

Turns roughly ten form entries into a print-ready bulletin and an 80-slide deck. In continuous weekly production use since March 2026.

LiveReact 19FastAPI

Replaces hours of manual formatting every week

OpenSteward

Offline-first ledger where two people on two machines converge without a server, a lock, or agreeing what time it is.

In developmentHybrid Logical Clock

Event-sourced, append-only, deterministic convergence

Q&AI

A deployed inference engine that narrows ~990 entities to one in 25 questions or fewer, selecting each question by expected information gain.

LiveTypeScript

Full-corpus regression harness gates every deploy at 98%

h@ck3r

A 1992 machine simulated in a browser: BIOS, virtual filesystem, shell, a second bootable OS, and a working BASIC interpreter.

In developmentTypeScriptTauri

808-line interpreter with a cooperative scheduler

OpenVoice

A self-hosted replacement for a commercial speech vendor, including a multi-stage voice-training pipeline that runs as a monitored background job.

DeployedPyTorchLoRA

Two engines, isolated environments, one API

Encounter

A desktop combat tracker shipped through 13 releases in four months, every one of them driven by a real user at a real table.

In useReact 19Tauri

The longest sustained release cadence in the portfolio

OpenSign

Digital signage a volunteer can run from any browser on the network, serving images in place from wherever they already live.

In developmentFastAPIReact

One port in dev and prod, on purpose

How I build

The part that doesn't fit on a resume

I direct AI coding agents, and I own the architecture, the constraints, the verification, and the release gate. My engineering background is what lets me specify and review what the agents produce; the years I spent away from the keyboard running organizations are what let me give up authorship without feeling like I'd lost something. The judgment is the job. What to build, what it must satisfy, and what stops it from shipping broken.

That only works if verification is real. Every claim on this site is checkable, and most of them are checkable by a machine — because I built the machine that checks them.

A governance framework for AI agents

~1,700 lines · ~20 concurrent projects · evolved through a documented incident log

Running agents across twenty projects surfaced a problem the tools don't solve: instructions that live in a session don't survive the session. A core safety rule dropped out of context once during a compaction, and a faithful re-read couldn't recover it — because the rule wasn't written anywhere durable. So I built the thing that was missing.

Background

Engineer, then executive, then both

I spent seven years as a software engineer and team lead — Tivoli Systems (IBM), where I was promoted to lead a team of eight; Interactive Intelligence, reporting to the company president; BroadJump/Motive, where I was primary architect of an automated build-verification system that was continuous integration in substance five years before the term became standard; Sun Microsystems; and BuildForge, a DevOps startup later acquired by IBM. I was also one of three founders of an early web company.

Then I left to become a pastor, and spent twenty-one years leading organizations — hiring and developing staff, owning operating budgets, directing a $500,000 capital project completed debt-free, and standing up every week to explain difficult things to people who did not come pre-equipped. I earned a doctorate in that last part.

The thread through all of it is the same one: find the work a person is doing by hand, and make it stop being necessary. That was process automation in 2001. It's AI-directed delivery now. The instrument changed; the job didn't.