A Pokémon card identifier, built for fun
A side project that answered a client question: how far can AI go when you hand it a photo and ask what it is looking at?
The problem
Clients kept asking whether AI could read images well enough to be useful in real work. Rather than answer in theory, we picked a problem with thousands of near-identical variations, where the differences that matter are small and the answer is easy to check: identifying a Pokémon card from a photo.
What AI did
The AI reads a photo of a card, picks out the name, set symbol, card number and rarity marks, and matches them against a reference list. It handles the repetitive part: thousands of cards, tiny printed details, and the same job over and over.
What people did
We built the reference data, decided which details the tool should trust and which it should double-check, and tested it against cards we already knew. When it was wrong, we worked out why and changed the instructions, not the answer.
A working identifier, and a clear picture of what image recognition does well and where it still needs a person. That picture now shapes how we recommend AI for product catalogs, inventory and anything else that starts with a photo.
Capability is cheap now. Judgement is what you're paying for.
Tell us where AI is helping your marketing and where it's making you nervous. We'll tell you what we'd keep, what we'd change and what we'd stop.
Book a strategy call