BUILD SYSTEMS · REAL PROBLEMS · 2026
AI / SYSTEM)FOR REALPROBLEMS.FEATURED PROJECTS
Not a repository dump — these are projects with external signal, active maintenance, deep investment or a distinct idea.
My current flagship: a self-hosted AI office that brings knowledge, agents, permissions, automation and enterprise tools into one maintainable system.
A minimal OpenCore EFI for Haswell + H81 + HD4600 + ALC662, and currently the public project with the strongest external GitHub signal.
Rust CLI/MCP plus a skill rule layer that helps AI agents discover mature tools and prior experience before writing incidental code.
A general-purpose AI-writing audit and full-polish skill with traceable provenance, mature upstream methods, and native Chinese, B2B and SEO/GEO rules.
A deterministic shared-memory architecture explored for Codex, Claude Code and Hermes Agent, representing a substantial multi-agent memory engineering effort.
An interactive tool that turns lottery combinations, cost and return into a computable model. Its lasting value is the idea, modeling and product thinking.
A macOS menu bar utility that uses Memory Pressure, Swap and background state to safely quit eligible apps under explicit rules.
A lightweight utility for safely disabling and restoring automatic macOS system updates — focused, low-dependency and intentionally practical.
REUSABLE KNOW-HOW
Reusable instructions from real design and engineering work, ready for AI agents to follow and validate.
Help AI agents create optically layered Liquid Glass interfaces instead of simply blurring a background.
A GSAP workflow for reversible scroll motion: choreograph by content meaning and keep content visible when scrolling stops.
JOURNEY
The direction changes; the through-line stays the same: reduce repetition, friction and unnecessary complexity.
From Gin, WebSocket, Redis and crawling to microservices and small business systems — the backend and systems foundation.
Code generation, asset processing, upgrade tools and automation scripts that compress repetitive workflows.
From skills, agent tools and knowledge systems toward permissions, workflows, maintainable deployment and complete enterprise AI systems.
PRINCIPLES
Technology is a means. A good system should make the work simpler, not make people adapt to the system.
Start from an actual need or friction point, not from a technology looking for a problem.
Reuse mature capabilities when possible; if a smaller solution is enough, don't introduce a larger system.
Deployment, upgrades, rollback and long-term maintenance are part of the product, not afterthoughts.
