Software engineer building durable AI agent runtimes, desktop applications, and developer platforms.
TypeScript · React · Node.js · Electron · PostgreSQL · LangChain
- Designing durable ReAct runtimes with recovery, checkpointing, auditable tool execution, and human-in-the-loop workflows.
- Building desktop and web experiences that make long-running agent work understandable and controllable.
- Improving context engineering, token budgeting, runtime observability, and reliable model integration.
| Project | What it explores |
|---|---|
| Agent Runtime v2 | A durable TypeScript ReAct runtime with PostgreSQL persistence, recoverable jobs, LangChain model integration, Plans as tools, and an authoritative event timeline. |
| Agent Code UI | A React 19 agent workspace with rich Markdown, tool activity, WebContainer-backed code execution, and inspectable runtime state. |
| Code Agent | A business-decoupled ReAct agent core built around LangChain, LangGraph, NestJS adapters, and extensible tools. |
| Infinite View | Infinite-canvas interaction experiments using React, Moveable, Infinite Viewer, and Zustand. |
| Weapp Components | Reusable lottery interaction components for WeChat Mini Programs. |
| Wordflow | Structured rich-text editing experiments built with Lexical and React. |
AI agent architecture Durable execution Context engineering
Developer tooling Desktop applications Web platform engineering
Runtime observability Human-in-the-loop UX Reliable model integration
I prefer systems whose behavior can be traced from user intent to persisted state, model calls, tool execution, and the final interface. My current work combines product-facing UX with the less visible runtime concerns that make agent software dependable.
Building systems that stay understandable after the demo ends.