Open to full-time and contract work · Taipei
Jean Shien — across product and engineering, and I ship what I spec
An engineering mindset, so what AI builds is more than a shiny prototype
PM and design experience, so every decision weighs its cost and boundaries first
Tools and technologies I work with
ABOUT
UI/UX taught me to read people; the PM years taught me to speak engineering. Now I use AI to connect both ends: I scope the requirements, trace the root causes, and write the tools myself — verified against real data, with the decisions written down. I never take an AI answer at face value. Every conversation feeds the business context behind the code and the data back into it, so the next answer comes faster and lands closer.
1,702 MCP calls, a 22× query speedup, 724 unit tests — every number measured in production, not estimated on a slide.
CAPABILITIES
From rule design to code — turning business logic into tools the whole company shares.
Breaking "I want a feature" into boundaries, states, edge cases and definitions.
Query Store, T-SQL, Mongo logs — evidence over intuition.
Clickable HTML prototypes to run the spec meeting; a full test matrix to guard the release.
Chat layer → skill layer → tool layer, so individual know-how becomes a team asset.
Pick the evidence, state the definitions, write the ADR — decisions that hold up when revisited.
SELECTED WORK
Fresh-grocery e-commerce · Aug 2026 – present
Locking "how this company should actually query" into 12 read-only tools — so ops ask their own AI and get the answers they used to queue for.
Read the case →Fresh-grocery e-commerce · May – Aug 2026
A PM who doesn't write the code, using database evidence to pin each root cause and staging 21 fixes across three waves — checkout SQL ended up 20,000× faster.
Read the case →Fresh-grocery e-commerce · May – Jul 2026
As PM I blocked the all-AI proposal — customers scan a QR and order themselves, staff import in one click. One form removed the entire problem.
Read the case →Fresh-grocery e-commerce · Jul – Sep 2026
Breaking the usual build pattern: the rules aren't hard-coded in if-else — when the AI isn't sure about an order it stops and asks a human, and the answer becomes a rule.
Read the case →Fresh-grocery e-commerce · Mar 2026 – present
An AI bot living inside Teams: searches three knowledge sources at once and always cites its source — making company knowledge faster to look up than to ask for.
Read the case →EXPERIENCE
Fresh-grocery e-commerce · 30-person team · Taipei
Greenbox Agricultural Enterprise Co., Ltd.｜Fresh-grocery e-commerce｜Managed 5–8 people
Yalu Technology · SaaS products
Shuoben Trading · e-commerce visuals
EDUCATION
LANGUAGES
LET'S TALK
AI tooling, system performance work, or turning a vague requirement into something that runs in production — happy to talk.