I fight the speed of light with software.
I go after problems where milliseconds decide the outcome and the result lands in the real world: market data that arrives before the trade settles, models that run entirely on the phone in your pocket, game worlds that react inside a single frame.
Photographer turned engineer. Scroll and I'll walk you through the last six years.
I spent years photographing live music on the East Coast. I worked as a house photographer for music venues, including Franklin Music Hall in Philadelphia and I.M.P.'s venues in Washington, D.C., and that work led to shooting for the artists themselves.
Living and working in Baltimore, I shot the streets, including the 12 O'Clock Boys, the city's dirt bike riders. Catching a wheelie at its peak is a lesson in timing I still use every day.
I wanted to sell my photography online before there was any plug-and-play way to do it, so I learned smart contract development. I wrote and deployed contracts, built the minting sites around them and shipped tools like rarity rankings and sales bots. That's how I got into software.
Camera Person was a collective art collection and gallery: 122 photographers from 39 countries in one collection, with every sale split equally on-chain. It was the first time I saw how much people can build together online when software lets them work across borders.
As a photographer, Stable Diffusion felt like a new kind of camera. I trained custom models on my own face, on the people I photographed and on specific subjects, like the yellow Spot robot dog, then used inpainting to place them precisely inside a scene. From stills I moved into motion and video, which led to producing AI video for an AMC documentary.
I freelanced and shipped my own AI products, and sold two of them: DoFDGPT, which turns dense government contract opportunities into clear, actionable briefs, and SnapCycle, an AI recycling assistant.
Mithril runs language models entirely on iPhone and iPad: Apple's on-device model plus open GGUF models from Hugging Face, a memory-fit check before every download and a live engine dashboard (tokens per second, time to first token, cache).
It acts with permission through Reminders, Calendar, Messages, Maps and Shortcuts, transcribes on-device with NVIDIA Parakeet and maps every network request. 4.5★ on the App Store. It also pulled me into the runtimes underneath: I now contribute regularly to llama.cpp, mostly reranker and encoder model support.
I built Polynode, real-time data infrastructure for prediction markets. It decodes orders 3–5 seconds before they land onchain and streams them to the trading products built on top. agg.market acquired it in July 2026.
Before that, Quantish let AI agents trade prediction markets and moved about $500K in volume.
Since the acquisition I've kept working with the agg.market team across the product. The highlight so far is its native app, built in Swift and SwiftUI: every prediction market venue on one screen, smart order routing that splits each trade for the best price, and live crypto rounds, on iPhone, iPad and Mac from one codebase. Now in public beta on TestFlight.
Any world. Any gameplay. Any character.
MASHBOX puts the worlds and play styles of the games you already own into one engine. Skate through portals in Halo and Stormwind, web-swing over Los Santos, play as Master Chief or Spider-Man and switch styles with the D-pad, with multiplayer. The Windows app is out now in free early access, and our own worlds run right in the browser.
Return of the Colossi: a Shadow of the Colossus homage, first playable in 48 hours with AI and open-source tools, now live in the browser.
Resident Evil 4: decompiled and modded in 12 hours, then running in the browser on WebAssembly and natively on ARM64 with no emulator.
I'm looking for the hardest problems I can find, especially where speed meets the real world. Here's more of the work.
▼ CONTINUEAny world. Any gameplay. Any character. MASHBOX mixes the games you own into one game, with multiplayer. Free early access on Windows, with our own worlds playable in the browser.
mashbox.gg ▸Game names are trademarks of their owners. MASHBOX is not affiliated with them.
A browser game homage to Shadow of the Colossus. The first playable build came together in 48 hours, built with AI, three.js, WebGPU, Jolt physics and headless Blender.
Play it ▸
Oct 2026
I just decompiled and modded Resident Evil 4 (PC) in 12 hours. People have been working on modding this game for 10+ years.
316 likes · 122 bookmarks
Oct 2026
Resident Evil 4 played as enemy characters, and running in the browser on WASM and on ARM64 with no emulators.
Sep 2026
A generative take on Shadow of the Colossus in 48 hours, and it's actually fun to play.
Sep 2025
Run LLMs 100% locally on your iPhone. Mithril, a free suite for local models and runtimes.
9.2K views
Sep 2025
Infipedia: a truly infinite Wikipedia that runs on local LLMs, entirely on your iPhone, even in airplane mode.
Jun 2025
Turned my MRI scan from a CD into a 3D model of my cerebral veins with a few scripts.
May 2025
Built a browser agent that applied to a software engineering job for me.
Feb 2024
Stable Diffusion animation for a TV docu-series: generate with a specific model, then add motion.
Sep 2023
None of these portraits are real photographs. Each comes from a Stable Diffusion model trained on my photos of that person.
