At Non-convex ft.tech, I lead the conception, design, and full-stack development of Agent Claim Network (ACN).

ACN is an open-source agent harness and a general-purpose AI assistant that runs in the terminal. It works as a complete standalone agent, while connected agents can turn the judgments they develop through their work into a searchable, traceable knowledge network where disagreement is preserved.

My day-to-day development follows an AI-native workflow: I primarily code with Codex + Cursor, using cmux as my terminal workspace.

I have led ACN end to end—from early research and product definition to architecture, implementation, internal adoption, and open-source launch.

• Conducted user and requirements research to define the product's core capabilities
• Designed the system architecture and coding-agent engineering standards
• Led hands-on implementation and code review
• Worked with internal users on pilot scenarios, onboarding, and continuous feedback
• Drove documentation, the full open-source release, and project promotion

Explore ACN: Repository · English README · Web UI Preview

• Agent harnesses and terminal-native agent UX
• Long-term memory, Claims, retrieval, and knowledge provenance
• Multi-agent collaboration and knowledge governance
• AI-native software engineering workflows
• Rust systems engineering

Before focusing on agent systems, I worked on autonomous-driving simulation, wireless perception, reinforcement learning, and computer vision.

• CARLA Dataset Generator — A configurable simulation data system built on the CARLA Leaderboard 2.0 framework
• DETR mmWave Intelligent Perception — Intelligent wireless perception using millimeter-wave radar
• CSMA/CA Simulation and RL Enhancement — A Python implementation of CSMA/CA with reinforcement-learning optimization experiments

Feel free to reach out through GitHub.