Operator's judgment.
Investor's rigor.
I'm Li — a software architect and AI-systems builder in Fairfax, Virginia. I lead with what I can ship: real, multi-agent systems that write, narrate, and support investment decisions.
I spent years as an engineering director leading 50+ engineers, with a full-stack background across Java, Android, and Python automation. Today I build production AI pipelines and orchestrate LLMs into systems that do durable work.
I think like a value investor: build slowly, verify ruthlessly, and let the proof speak. I'm a lifelong learner and a bilingual writer — 中文 and English — on engineering, AI, and decision-making.
From leading teams to building systems.
AI architect & independent builder
Self-directed · advisory
Design and ship multi-agent AI pipelines end-to-end — the editor, fairytale, and investment systems on this site. Advise teams on AI architecture and where automation actually pays off.
- LLM orchestration
- Multi-agent design
- Eval harnesses
- Advisory
Engineering Director
Led 50+ engineers
Owned delivery across multiple teams — hiring, architecture, and the judgment calls that don't fit in a ticket. Learned that the hardest part of engineering is deciding what not to build.
- Org leadership
- Architecture
- Delivery
- Hiring
Full-stack engineer
Java · Android · Python
Built and shipped across the stack — mobile, services, and the automation glue in between. The habits formed here — test it, measure it, version it — are the same ones in every pipeline I build now.
- Java
- Android
- Python automation
- Backend
What I build with.
Languages
- Python
- Java
- JavaScript
- SQL
AI / LLM
- LLM orchestration
- Multi-agent design
- Eval harnesses
- RAG & tools
Media pipelines
- ElevenLabs (voice)
- Gemini (image)
- Audio mixing
- EPUB / KDP
Platforms
- Android
- Backend services
- Automation
- Static web
How I make decisions.
Proof over pitch
Every project here carries a working artifact, not a claim. If I can't show it, I haven't really built it yet.
Verify ruthlessly
Test-driven, with held-out evals and deterministic checks. "It seems to work" is where the real work starts, not ends.
Judgment as a system
Good decisions should be repeatable. I encode frameworks so the same discipline applies on a calm day and a loud one.
Keep the human
Whether it's a writer's voice or an investor's call, the system serves the person. The tool is loud; the judgment has to be louder.
Let's build
something real.
Advisory on AI architecture and multi-agent systems — or just to compare notes. No hard sell.
- itlipan@gmail.com
- GitHub
- github.com/devlipan ↗
- Writing
- EN / 中文 essays