Co-founder, Ouro Labs · Core Maintainer, Ouroboros · Hansung University

I build tools and protocols for human–AI collaboration.

My work asks a practical question: how can AI make human judgment easier to examine, test, and act on while keeping people responsible for their decisions?

I am a core maintainer of Ouroboros, an open-source agent OS for AI-assisted software development. I work on specifying tasks, coordinating agent execution, and preserving evidence that people can use to evaluate the results.

Through Ouro Labs, a collaborative team, I also work on research and applied AI projects.

My merged contributions

A local-first experimental tool for recording decisions as structured DecisionTrace records, revisiting their original reasoning, and producing evidence-grounded runtime analysis.

An exploratory simulation of fictional human perspectives. It examines how the same situation can lead to different actions as body, memory, identity, and world-model conditions change. It is a thought experiment, not a validated model of real people.

• Pullim: Earlier protocol research, now also an experimental decision-support app. It helps turn a difficult situation into a clearer problem, usable evidence, a provisional decision, and a next action chosen by the person.
• 말결 (Malgyeol), with SergioBuilds: Bounded agent-payment infrastructure that connects human intent with policy checks, consent, purchasing, and payment evidence.
• FlashPatch, with SergioBuilds: Open-source visual safety QA for detecting potentially hazardous flashing, applying scoped repairs, rerunning checks, and recording the evidence.

I also contribute fixes and regression tests across the open-source ecosystem:

• Gajae Code · ⭐ 3.1k: provider cancellation and stream cleanup, CLI argument handling, and session statistics. (GitLab request controls, Ollama reader cleanup, CLI arguments, merged contributions)
• Oh My ClaudeCode (OMC) · ⭐ 40.6k: validation of context-threshold configuration, preventing malformed overrides from silently taking effect. (Merged PR)
• Oh My Codex (OMX) · ⭐ 33.7k: stripping OSC terminal-control sequences from notification text. (Merged PR)
• OpenCodex · ⭐ 18.2k: portable Bun test batching for macOS environments without GNU timeout. (Merged PR)
• SQLFluff · ⭐ 10.0k: Hive GROUP BY support for trailing WITH ROLLUP and WITH CUBE. (Merged PR)
• ThinkJS · ⭐ 5.3k: fixes for request-validation method leakage and shared router defaults.
• user-scanner · ⭐ 5.6k: consistent validator resolution between library and CLI modes. (Merged PR)

Star counts are approximate snapshots from 2026-11-04 and may change. These examples are merged upstream contributions; release and deployment status may differ.

codbeing records and revisits judgment. Pullim explores how judgment leads to action. 30p examines how choices can branch under different human conditions. Ouroboros provides an execution and verification layer for carrying out AI-assisted work with traceable evidence.

Across these projects, I am interested in systems where AI helps people see, challenge, and improve their reasoning while they remain accountable for the decisions.