We built an entire fake world with AI. Here's how

A year ago, building a believable, end-to-end fake food-delivery app as a solo project would have been absurd. The restaurants alone — names, cuisines, menus, prices, photos, for dozens of cities — would have been months of work and a photographer's budget. Today it's a weekend and an orchestration problem.

The world is the product

NomNomNever isn't a single fake restaurant. It's a whole city grid of them: real Korean convenience-store items with Won prices pulled from the actual store apps, chain menus with hidden secret items, local favorites per city, dish-matched food photography, and delivery riders with names and personalities. Every layer had to feel real enough to trigger a genuine craving — because if it doesn't feel real, it doesn't work.

AI as the whole studio

Menus, descriptions, and translations are generated and then hand-curated. Food photography is AI-generated and matched to specific dishes so the ramyeon looks like ramyeon and the jokbal looks like jokbal. Courier characters, city street names, the map, the tracking animation, the support-chat dialogue — all assembled with tools that a single person can now direct.

The build itself is the story: not "an AI made an app," but that the right orchestration of AI lets one person ship a world that used to require a team, a studio, and a warehouse of stock photos.

Convincing enough to satisfy a real craving

That's the bar we set — and the whole point. A fake world only helps if it's good enough to fool the part of your brain that wanted to order. Six months ago, hitting that bar solo wasn't possible. Now it is.