I build tools and protocols for human–AI collaboration.
My work asks a practical question: how can AI make human judgment more visible, testable, and actionable without replacing the person who has to live with the decision?
Maintainer work on an agent OS for AI-assisted software development: specifying work, coordinating execution, and preserving evidence that the result is safe to trust.
A local-first research tool for recording human decisions as structured DecisionTrace records and reading them back with evidence-grounded runtime analysis.
A decision-making protocol for turning difficult situations into a clear problem, usable evidence, a temporary decision, and one next action. The AI does not decide for the person; it helps make the decision inspectable.
An exploratory simulation of plural human judgment. It studies how the same stimulus can produce different actions across changing combinations of body, memory, world-model, and unconscious context.
- 말결 — exploring how language, relational context, and AI-mediated interaction shape one another.
- FlashPatch — open-source visual safety QA that detects, repairs, rechecks, and records evidence for dangerous flashing in games and video.
codbeing observes judgment. Pullim helps move judgment toward action. 30p examines how judgment branches across plural human conditions. Ouroboros is the operating and verification layer that lets this kind of AI-assisted work be carried out with durable evidence.
I am interested in systems where humans remain accountable for their decisions while AI makes the reasoning process easier to see, challenge, and improve.