I tried to build a modern, AI-powered book of cat fortune-telling. It got one person a win.
LuckyCat looks like a joke: upload your cat, get lottery numbers. Underneath it was a genuinely strange idea — could crowd data rebuild an ancient folk art, and tell us which cat is actually the luckiest?
Are cats lucky? Everyone in Taiwan grows up assuming so: cats bring money. The maneki-neko — the lucky cat, the beckoning cat — sits by every cash register. There are lottery-shop legends: a scratch card, left under a sleeping cat, that turned out to be the jackpot. The cat as a tiny, furry fortune engine. I never questioned it. I just wondered about the next step.
Which cat, exactly? Whose is the lucky cat — and can you actually prove it?
If you put every cat on the same baseline — the same lottery, the same numbers, enough draws — could you eventually see it? Which face is luckiest. Which coat. A physiognomy for cats, built not from folklore but from data.
That was the actual idea hiding under a web toy. Chinese folk tradition has a whole genre for this: xiang mao jing, the classic "art of reading cats" — the same intuition as face-reading for people, that appearance and fortune are somehow linked. I wanted to feed that old book into the most modern thing I could reach and see if a crowd, over time, would turn superstition into a statistic. An AI-era book of cat fortune-telling. That was the dream. LuckyCat was just the doorway.
01Which cat is the luckiest? Why the question wasn't (only) a joke
People assume anything with cats and lottery numbers is a gag. Fine. But look at the ground it was standing on. In 2025, Taiwan registered about 257,000 pets — 2.4 times the 108,000 babies born that year. Over a decade, registered cats exploded by roughly 327% while newborns nearly halved. We have quietly become an island of cat people, and a real "cat economy" has grown up around it.
So no, wanting to show off your cat, and wanting your cat to be special, is not a niche. It's most of a generation. The market was real; the affection was real. Somebody was going to build the shelf for all of that. I just wanted it to be a slightly stranger shelf than usual.
And the questions people actually ask are wonderfully specific. Is an orange cat lucky? Are black cats unlucky, or is that a Western import? Which coat brings the most money — calico, tabby, tuxedo? Folklore has opinions on all of it. What folklore has never had is a scoreboard. That was the gap.
And yes — someone will say a leaderboard of cats is objectifying animals. I hear it. All I can say is: I know my own intention, and that's enough for me. It's a fond, silly statistic about creatures people adore. Nobody accuses face-reading of being cruel when it's flattering.
02The patient zeros: Jigula and Maluchan
Every experiment needs its patient zero. Mine came in a pair. When I was first sketching this, there were two orange cats a friend and I had gone to adopt together — Jigula and Maluchan. They were the first faces uploaded, the very first test data. If any cats got this whole strange idea moving, it was these two.
Later I ended up with two of my own, a set of white brothers, and gave them names that were really wishes: Pangfu and Pangji — roughly "chubby fortune" and "chubby luck." That they'd grow up plump, healthy, blessed, and auspicious. They joined the database soon enough.
And here is the one law of this entire project I can state with total confidence, no dataset required: your own cat is always the cutest, the most beautiful, and the luckiest. The great book of cat fortune-telling will simply have to work around that.
03Does the cat actually pick the lottery numbers? (the honest answer)
Here's the confession the shelf label won't give you. The recognition — is this even a cat? — ran in the browser, because I refused to pay for anything, which meant stitching together free APIs that failed to connect roughly as often as they worked. When it worked, it was fine.
The numbers themselves? I did feed the old xiang mao jing into an AI and asked it to derive a modular set of coefficients — a way for a cat's features to nudge the draw. It was not good. It was a very long way from anything you could call analysis. So let me be precise about how much your cat participates in choosing your lottery numbers:
Ninety percent random. Ten percent my own special sauce. And the ten percent was mostly hope.
The grand physiognomy engine — classify every coat, run every face against outcomes, let the crowd surface the truly lucky patterns — was always the ideal state, never the shipped one. It never left friends-and-family testing. Which is a shame, because with enough people, it genuinely might have worked.
04The one win, delivered by hand
This is my favourite part, and the most LuckyCat thing that ever happened. During friends-and-family testing, one person's cat actually picked a winning ticket. An actual, real win.
And the automatic prize-checking system — the whole reason the thing was supposed to exist — wasn't finished yet. So I did what any world-class lottery platform does. I checked it myself, opened a messaging app, and typed the good news to my friend by hand. The luckiest cat on Earth, and its prize was delivered over chat like a text from your mum.
That's the status line you see on the shelf: 0 Winners. Not because no cat ever won — one did — but because the system was never awake to count it.
05Why it stopped: eaten by its own scope
I want to be honest about the ending, because it's the most common ending there is. I didn't hit a wall. I built a beautiful hallway and kept adding doors until I couldn't reach the exit.
The functions I wrote were good — genuinely. But I couldn't give it design, and design is the one thing the AI couldn't hand me either. Good features wrapped in something that looked unfinished. That stung. And then, instead of closing the loop, I kept widening it: finish the auto prize-check, add a prize-money leaderboard, then collect emails to notify winners, then send a weekly "luckiest cat of the week" EDM to everyone who joined… the scope grew every time I thought about it, until thinking about it was all I did. Then it just stopped.
Most projects don't die. They get quietly buried under the next three features you were sure you needed.
Looking back, the move was obvious: ship the complete flow — upload, automatic prize-check, prize-money and probability leaderboard — even if it only ever worked for the Taiwan Lotto. One country, done, is worth more than the whole planet, imagined. Whether it ever reaches every lottery on Earth — that part, I've made my peace with. That part can be up to fate.
Pangfu and Pangji are, as of this writing, still the two luckiest cats in a database almost nobody has seen. Which sounds about right. The luckiest cats in the world, waiting patiently for the world to notice — plump, healthy, and, obviously, the most beautiful.
Next field note, coming soon: The hardest part was never the cats — it was the automatic prize-checking. A post about the boring plumbing that quietly kills side projects.