LLM inference engine
2026 · C++ · Vulkan · AI
Runs large mixture-of-experts coding models on ordinary laptop GPUs.
Game dev, white-hat hacker, ML systems engineer. It started around 2011 on a slow Core 2 Duo, taking games apart just to see how they worked, and I never stopped building. Everything I run is self-hosted, on hardware I own.
My first computer was a Core 2 Duo with 4 GB of RAM. I spent it taking browser and PC games apart, editing memory to see what each number did. That is where I learned to read a system instead of just using it.
Breaking games turned into the better question: how do they defend themselves? That pulled me into memory, processes and the kernel underneath, and later into protecting my own games.
Taught myself to code by reading, studying and testing everything. Wrote code on my phone in Termux, modded apps and games, and built my first real apps.
Founded a game studio as CEO and built games for companies in China until 2018. We shipped in Unity, Unreal Engine and Construct 3, mostly in C# and C++, with concept art from collaborators.
Started building automated trading bots and market signal engines. Backtesting, indicators, exchange APIs and a lot of arithmetic. Still running and improving them today.
Formed a white-hat security group with researchers from Russia, Korea, China, the Philippines and Thailand, working together to find and fix vulnerabilities.
Shipped a mobile arcade game under one of several small studios I have started, and began publishing on GitHub: scripts, trading tools and the projects I keep learning from.
Self-hosted sites, Go services, Flutter apps and a Vulkan LLM inference engine, all running on hardware I own. Plus the QA habit that keeps regressions from shipping.
2026 · C++ · Vulkan · AI
Runs large mixture-of-experts coding models on ordinary laptop GPUs.
2026 · Go · AI
An agent that plans, writes and tests code against a local model.
2026 · Go · SvelteKit · SQLite
A prediction market for local events, running on a single-board computer.