Frontend & AI-product engineer·Guangdong, China
I build AI products and learning-first engineering projects.
Turning complex ideas into small, traceable steps — framework internals, build tools, compilers, and practical AI applications. A running record kept in public, mostly on GitHub, occasionally in essays.
“Talk is cheap. Show me the code.”
New here? Follow the journal on GitHub →
Products I'm shipping right now. Each one is live somewhere you can click; each one commits small.
An AI agent that reads your PDFs and draws editable mind maps. A visible multi-step tool-calling loop, built-in RAG (no LangChain), bring-your-own-key, multi-provider — the reasoning is legible, the output is yours to keep.
Previewable Flutter animations — real running Flutter web, not GIFs — each one paired with its traceable pitfalls and a parameter playground. Reachable from Claude Code or Cursor as a remote MCP tool.
A multi-provider AI image playground — text-to-image, edit, inpaint. Entirely client-side: bring your own API key; nothing ever touches my server.
Convert an API's documentation into a working MCP service, so AI coding IDEs — Claude Code, Cursor — can call your APIs directly, without you writing any plumbing.
Reads a 1,000-year-old system as a data problem: it renders a person's 八字 (BaZi) as a stock-style K-line chart. Enter a birth date and it plots a life's fortunes as candlesticks, with AI-written readings across character, career, wealth, and health.
A deliberately minimal tennis scorer — offline-first and installable, with no account and no network required. Games, sets, and tiebreaks are tracked with state kept entirely on-device, so it never drops a point courtside.
Work landed in projects I use every day — small, defensible patches to codebases many multiples my size.
The best way I know to understand a tool is to write a small, honest version of it. Each of these is short enough to read in one sitting and complete enough to run.
↳github.com/xianjianlf2/<name>