English · 简体中文
AI Builder · Forward-Deployed Engineering · Agent Systems
I turn ambiguous business workflows into working AI-native products — from product framing and agent architecture to deployment, diagnostics, and iteration.
我把模糊的业务流程，做成可部署、可运维、可持续迭代的 AI 产品。
| Project | What it demonstrates | Focus |
|---|---|---|
| Agent Runtime Map | An evidence-backed analyzer and circuit-style viewer for agent codebases. It builds deterministic topology from code, keeps file:line provenance, and supports route inspection, static playback, and diagnostics. |
TypeScript / JavaScript / Python · AST analysis · React Flow · ELK · CLI |
| Commercial Invoice Generator | A workflow product that turns a proforma invoice, export license, and booking note into a print-ready commercial invoice, with explicit privacy boundaries and a deployable web app. Live demo. | Python · Flask · PyMuPDF · HTML-to-PDF · Docker |
| Personal Site | A trilingual portfolio and product surface with an AI chat experience backed by a Netlify Edge Function, keeping the provider token off the client. Live site. | HTML / CSS / JavaScript · Deno · Netlify Edge Functions · Coze API |
- Tools that make agent behavior inspectable, evidence-backed, and easier to debug.
- AI and automation products for real operational workflows, especially where reliability and handoff matter.
- Complete product systems: interface, workflow, deployment, diagnostics, and the iteration loop after launch.
- Grounded agent tooling: Agent Runtime Map separates deterministic code analysis from optional semantic enrichment; the LLM layer can describe existing structure but cannot invent topology.
- Operational automation: Invoice Generator documents its parsing pipeline, temporary-file handling, browser-side stamp storage, Docker deployment path, and 32 unit tests.
- Deployed product surfaces: the invoice workflow and personal site are available as live products, not only screenshots or design files.
- Engineering transparency: project documentation calls out scope, architecture, known limits, privacy behavior, and reproducible setup.
- 我用 AI 做了这么多产品后，终于搞懂 Agent 到底是什么
Workflow、Tool、Runtime、Harness，以及 Agent 与 Workflow 真正的分界线：到底是谁在决定下一步做什么。
I write technical documentation alongside the code, with an emphasis on design decisions, evidence, and operational limits.
- Agent Runtime Map — Architecture
- Project Context & Semantic Analysis
- Reusable Visual Components
- Invoice Generator — product and deployment notes
Agent systems · AI-native product delivery · Static analysis and developer tooling · Enterprise workflow automation · Full-stack prototyping
Working across TypeScript, JavaScript, Python, React, Flask, Deno, Docker, and serverless deployment — choosing the smallest stack that can ship and stay understandable.
Interested in agent infrastructure, AI-native product delivery, and forward-deployed engineering.