Preplo turns cooking videos into recipes you can cook

People save cooking videos and rarely cook them. Short videos are good at inspiration and bad at instructions: quantities flash past, steps get skipped, and cooking from one means pausing, rewinding and guessing.

The contest brief came from a cooking creator and named this gap: help people go from "I saw this" to "it's on the table". RevenueCat gave builders four weeks to build, launch and monetize a working app.

Two things made it hard to build. YouTube, TikTok and Instagram expose recipes in different ways, and some block cloud servers. And early AI output was not safe to cook from: missing ingredients, invented quantities, the wrong language. A recipe you cannot trust is a recipe nobody cooks.

We started from the moment the app has to serve: a saved video, a kitchen, messy hands. Then we cut scope hard. Meal planning, social feeds and community stayed out of the contest build so the core could be polished.

The model was the easy part. Making its output trustworthy was the work. We put a validation layer around it: strict output rules, unit standardization, translation and detection of videos that are not recipes. If a video has no recipe, Preplo says so instead of inventing one.

We designed the running costs up front. Every extraction is cached, so the same video is never analyzed twice, and recipe changes are cached too. Subscriptions, a free trial and premium access were in from day one, as the contest required.