Research & competition
Pokémon TCG
From experiments to better game decisions
Learning components and a separately tested matchup counter, developed through roughly two months of agent-assisted research.
Research & engineering
Selected work
I build tools, test ideas, and follow the questions that emerge. These projects show different parts of that work: learning systems, multi-model applications, and practical control of agent workflows.
Featured case studies
Multi-model application
The Council
Make the reasoning inspectable
An application that brings together answers, peer critique, claim checking, and synthesis across model providers.
Agent security
BigBoss
Gate the action, keep the record
A local approval gate and persistent decision store for AI coding agents: each proposed action is hashed, routed through policy, and decided by a person. A personal working MVP; enforcement depends on the adapter.
Research areas
Several of these projects answer to one question: how a person keeps authority over what an AI agent does, and what record is left behind. Agent security research →
Methods and explorations
Smaller tools and research directions that connect to the larger program.
SEED →
A measurement protocol and checking tools for agent-driven work. The proposed outcome benefit remains a research question.
Godot AI Methodology →
Public methods and checks for making game-code architecture easier to inspect and modify with AI assistance.
The Bus →
A retired coordination exploration with its design history and retraction record preserved.
Follow the evidence
The case studies explain what I contributed and what the evidence shows. For a deeper look, explore the Pokémon evidence map or the agent-workflow census. Experiments, implementation, and future objectives have different kinds of evidence.
AI agents assist my implementation, research, and writing. Each case credits the public tools and implementations involved.
Discuss a project or opportunity →