The longer version

I started writing code years before language models went mainstream, building Minecraft bots that could hold a conversation in chat. Back then that meant hand-rolled heuristics and state machines, not an API call. That side project turned into a long detour through web development, and eventually into machine learning, which I taught myself before I took it seriously.

In 2025 I finished the University of Utah’s machine learning bootcamp and earned their Artificial Intelligence certificate. I don’t have a computer science degree, and this site won’t pretend otherwise. Most of what I know I learned by building something and then finding out where it breaks.

Building

Then I shipped things. WrapsRL generated Rocket League car wraps with Gemini and made $600 in its first month; I’m shutting it down now that Easy Anti-Cheat closes the modding path it relied on. CanIRunAI tracks more than 1,284 open models and tells you which ones your hardware can run. Adversec is an API for red-teaming language models: prompt injection, jailbreaks, data leakage.

Each one taught me something the bootcamp couldn’t. Billing, queues, row-level security, rate limits, and the fact that “AI product” has to mean more than calling an API somewhere.

Worrying, usefully

Somewhere between those projects, my attention moved. Capability kept compounding. Open weights caught up with the frontier. At least one frontier lab now has its own model writing most of the code it merges. I took BlueDot Impact’s course and earned their Future of AI certification in 2026, and since then I’ve spent more of my time on the questions underneath the products: what these systems are for, who controls them, and what happens when they work better than anyone planned.

Red-teaming made that concrete. Adversec started as a security tool. Every probe it runs is the same experiment: change the input a little and see whether the output changes a lot. That’s not only a security question. It’s a question about whether we can predict what we’ve built.

I’m not worried for effect. I build with these tools every day and they’re the best tools I’ve ever had. That’s exactly why I take the risks seriously. The logs are where I work through it, in public, with the receipts.

Next

This fall I’m starting at ASU Online through the Starbucks College Achievement Plan, beginning with MAT 114 and CIS 105. The plan is to finish a degree while I keep working, and keep building while I study.

Outside of software I make music as Lambourne. I built UNVAULTED to keep my unreleased songs and demos in one place, and I started BTRFLY, a Rocket League tournament org.