Hong Kong · AI · maths · tennis · violin · food
I build AI products, help businesses decide where AI is useful, and teach people how to use it. Based in Hong Kong. Usually up for tennis or a good meal.
In 2002, I started an AI society in Hong Kong. It lasted less than two years. My interest in AI lasted rather longer.
So I went into finance instead, at AXA, Thomson Reuters and Telstra. It was a great education in how big companies actually make decisions.
In 2015 I came back to tech and co-founded WeMine. We built marketing technology for more than 100 enterprises, from WeChat and Christie’s to BMW and MTR Malls, and I learned what it really takes to grow a team and a company. In 2021 I went back to school for a master’s in data science at the University of Michigan.
Today I split my time between consulting, building and teaching. I still write a lot of the code myself, because that’s the part I enjoy most. My articles and classes are at HtPi (htpi.me).
A selection from the products and experiments I’ve built over the years, independently and with teams.
Social CRM and marketing automation used by more than 100 enterprises. It ran customer conversations on WeChat and WhatsApp for brands across Hong Kong.
Co-founder & CTO · Nov 2015–Jun 2023My part
Co-founder and CTO. Took the platform from a first prototype to use by more than 100 enterprises.
What it does
Customer conversations and marketing automation across WeChat and WhatsApp.
An AI assistant inside WeMine’s Slack workspace that helped the team find internal knowledge and handle everyday workflows.
LLM · RAGWhere it lives
Inside the team’s Slack workspace at WeMine.
What it helps with
Finding internal knowledge and handling everyday workflows, using LLMs and retrieval.
A food app that recommends restaurants by your own taste instead of star ratings, using machine learning to work out what you actually like.
Co-founder · 2019–23My part
Co-founder; led the product and data science teams.
The idea
Learn each person’s taste in food, rather than assume one restaurant rating fits everyone.
A real-time multiplayer poker platform I built on my own, with provably fair dealing and an AI that learns how each player bets.
Solo buildMy part
Built the platform on my own, from the game engine to the AI opponent.
What’s inside
Real-time multiplayer poker, provably fair dealing, and an AI that learns how each player bets.
Recent work also includes FundJournal, Food Partner and HtPi.
Consulting
I help you decide what’s worth building with AI, then build it. I’ve run a company and still write the code, so I can work through both the business decisions and the technical details.
Choose where AI can help, what to build first, and how to handle the data and risks.
AI agents and LLM systems, from the first prototype to something your customers actually use.
For companies that need a CTO, just not five days a week. Architecture, hiring, vendors and the big technical calls.
Hands-on AI sessions built around your team’s work, data and tools.
Not sure it’s a fit? Tell me what you’re working on and I’ll be honest about whether I can help. Fixed-price projects are fine too.
I teach AI and coding to students, professionals and teams. My teaching includes AI in Asset Management at the HKU Summer Institute, guest lectures at other universities, classes at Yew Chung International School, and my own courses.
I also used to coach students for regional maths olympiad contests, and I’d be glad to do it again.
Schools, companies and parents are all welcome to get in touch. Classes are priced separately from consulting.
Click any role for details.
Want my full CV? Email me and I’ll send the latest one, or find me on LinkedIn.
Some interviews and talks over the years, plus a few animations I made for HtPi.
Read more: Fortune Insight interview ↗ · HK01 on BudsAlike ↗
A few things I do when I’m away from the keyboard. Play a point, pluck a string, or test your poker instincts.
I’m always up for a game. Until then, you can play the virtual me. Move with your mouse, finger or the arrow keys, and tap or press Space to swing. Swing early to go cross-court, late to go down the line.
I play the violin. Try plucking a string here: tap to hear a note, or drag and release for a stronger pluck. On the shaded fingerboard, notes get higher as you move right. Sound on.
Choose G, D, A or E with ↑/↓ on the strings, choose a note with ←/→, then press Space to pluck. The buttons below also play notes. The tune uses a bowed-string-style synthesiser; free play uses a plucked-string model. Slow the tempo to follow the fingering.
I love poker because it’s maths with people in it. Quick test: two hands, all-in before the flop. Which one is ahead?
Estimated from 12,000 random boards, sharing tied pots equally. The percentage is equity, not a promise about the next deal. Near-equal hands count as a close call.
I like running. I also like eating. Here’s a different way to picture a week’s kilometres: the equivalent energy in something good to eat.
Approximate energy equivalents. Portions and individual needs vary.
This is an energy comparison, not an exact extra-food allowance. Meals vary; 700 kcal is an editable example, not a universal average. Food portions: wonton noodles 630 g / 447 kcal; egg tart 70 g / 217 kcal; pineapple bun 97 g / 349 kcal, from Hong Kong’s Centre for Health Protection.
Level-running estimates interpolate the 2024 Adult Compendium speed/MET values. Extra energy = (MET − 1) × 3.5 × weight in kg ÷ 200 × minutes, subtracting estimated resting energy during that time. Terrain, physiology and real portions vary. Your inputs stay in this page.
At secondary school I won awards in the IMO Preliminary Selection Contest – Hong Kong a few times, and each time went on to months of olympiad training. Later I coached students for regional contests.
I’ve wanted to prove Goldbach’s conjecture for years. So far, I’ve built this box.
Goldbach’s conjecture: every even number above 2 is the sum of two primes. A pair checks this example; a proof would explain why it always works. Explore whole even numbers from 4 to 100,000.
I love exploring food. I once compared a popular dining guide with places I’d actually eaten at, and the top-rated ones often weren’t that good. Taste is personal, so I co-built BudsAlike to recommend food by your own taste.
Tell me where to eat →I write in Chinese and English about AI, maths, probability and whatever I can’t stop thinking about that week. Most of it is at htpi.me, and some has appeared on company websites and in the press.