Toryo
A TypeScript coding-agent orchestrator with evidence-bound review, configured checks, and reversible Git checkpoints.
- 8 tool adapters
- Quality ratcheting
- Trust-based delegation
Selected work · The Workshop
Autonomous agents, orchestration infrastructure, research copilots, and games. The common thread is a preference for measurable behavior over polished claims.
Bring me a hard problem →Flagship case studies
3 systemsA TypeScript coding-agent orchestrator with evidence-bound review, configured checks, and reversible Git checkpoints.
Five local-LLM agents sharing a Minecraft world, with persistent memory, role-specific tools, and an experimental evaluation track.

A Python evaluation harness measuring fact loss across repeated summaries, with deterministic grading, committed transcripts, and corrected grader bugs.
More experiments
16 projectsDark-fantasy idle RPG with dungeon gates, spirit binding, and prestige systems
Real-time dashboard for AI agent teams with anime-styled command centers
Free idle RPG dungeon crawler with zero microtransactions
Multi-repo orchestrator that dispatches parallel Claude Code sessions autonomously
An execution engine for agent fleets, built on the premise that agents lie about finishing
Browser-based math game for children with multiple difficulty levels
AI-powered language learning with FSRS spaced repetition for Japanese, Spanish, and Portuguese
Domain-specific AI research copilots — 7 repos spanning math, law, quant, engineering, chemistry, biology, and policy
Systematic review of AI tutoring effectiveness in K-12 mathematics
Improved concentration bounds for chi(G(n, n^{-1/2}))
Percy Jackson Minecraft mod with celestial weapons, mythological monsters, and prophecy quests
Vision Transformer vs CNN for gravitational wave glitch classification
A JavaScript text game based on Lord of the Rings, built for my kids