Quantization
Round 64 synthetic weights to fewer bits and watch the error grow.
Uniform symmetric quantization of a synthetic weight vector. An illustration, not a benchmark.
I’m an independent developer and AI engineer. I build software, games and AI tooling: websites and developer tools, Discord bots and automation, and interactive game experiences.
Right now I’m developing Basketball Heroes, a multiplayer VR/XR basketball game in Unity, and studying AI engineering, machine learning and large language models. Much of what I make for other developers is open source: roblox-ai, a library of 12,000+ Roblox UI references for AI coding agents, has 221 GitHub stars.
A multiplayer VR/XR basketball game developed using Unity.

A curated library of 12,000+ Roblox UI references for Claude Code, Codex and other AI coding assistants to study as design inspiration.


A skill for Claude Code and Codex that helps them build modern, good-looking Roblox game interfaces.
Connects Discord to Claude Code and shows your live chat, AI model and token usage on your profile so people can see your project.
Basketball Heroes is a multiplayer VR/XR basketball game I’m developing and managing in Unity. Here are the ball skins from the game. Pick one to preview it.
Join the Discord

01 / 18Classic
I’m learning how language models work by taking them apart: tokenization, quantization, architecture and training, plus open-weight models running locally through Ollama. Two small experiments below run in your browser.
Round 64 synthetic weights to fewer bits and watch the error grow.
Uniform symmetric quantization of a synthetic weight vector. An illustration, not a benchmark.
A tiny byte-pair encoder, trained live in your browser.
Learned from a short built-in text, so common words merge into single tokens and unfamiliar ones fall apart. Real tokenizers train on vastly more data. A dot marks a leading space.
No results.