I build AI tools people can understand and control. I like making the work visible: what an agent is doing, what evidence supports its output, and how to recover when a change goes wrong.
Most of that work is in TypeScript, Python, and React. The projects below range from local developer tools to experiments and a math game.
Tenshu, running in simulated demo mode. These are sample agents and tasks, not a live connected system. View full-size screenshot.
- Toryo — Make coding-agent changes easier to review and undo. This TypeScript orchestrator ties review to captured evidence and configured deterministic checks, with recoverable Git checkpoints tested using real Git transactions. Checks run as trusted local code.
- Tenshu — See what an agent team is doing in one place. A local React/TypeScript dashboard combines typed agent and session views with a Hono API and WebSocket updates. Includes tests and the simulated demo shown above.
- Minecraft Agent Swarm — Explore how local agents coordinate in a shared world. Built with TypeScript, Ollama, and Mineflayer, with shared context, persistent memory, and telemetry. Its separate evaluation track is experimental; it has not demonstrated model learning.
- BattleMath — Give math practice a game loop. A React/TypeScript frontend with adaptive difficulty, hints, daily challenges, and progress saved locally.
Contact: jesseweigel@gmail.com