I build AI products from my own questions — and when they fail, I write down why.
By day I deploy LLMs across China's phone makers at Alibaba Cloud's Tongyi (Qwen); on the side I build my own tools and document what breaks.
What I'm building
VidSnapbuilding
An AI agent that turns any video into structured, illustrated notes.
Started Jan 2025. An agent that orchestrates skills — transcription, keyframe understanding, and downstream generation like illustrated notes — into one video-understanding harness. Still pre-launch; the current push is getting it in front of first real users.
YouTube Download—shut down
A paid YouTube downloader I shipped end-to-end — closed once the unit economics didn't work.
~50 monthly users · 1 paying customer
Built with a mentor from the Antalpha days. Full stack shipped: SEO, a live payment system, real traffic. It died on unit economics — no overseas entity for Stripe, and proxy/traffic costs outran revenue at ~50 monthly users and one paying customer. The most useful failure I've run.
Track record
2025 — now
Product Development Solution Architect
Alibaba Cloud · Tongyi (Qwen)
Deploying LLMs across every major phone maker in China.
2024
Live Caption Translation · Dell
Client project
On-device caption translation for Dell's AI PCs; demoed at Singapore FinTech Festival.
2024
AI Engineer Intern
HyperGAI
VLM inference & evaluation pipelines.
2023 — 24
AI Product Intern
Antalpha
Shipped cc.Monet to 180 staff; −90% weekly processing time.
2022
AI Engineer Intern
Procter & Gamble
CNN defect detection — 2% false pass vs 10% target.
2023 — 25
M.Eng, Materials Science + AI
National University of Singapore
Built an LLM pipeline over 41,000 papers; published in Chinese Physics B.
2019 — 23
B.Eng, Materials Science
South China University of Technology