Menu-Order AI: GLP-1 Meal Finder

Eating out on a specific nutrition plan means guessing. A menu tells you what a dish is called, not what is in it, and someone tracking protein or managing carbs has to either interrogate the server or accept the uncertainty.

GLP-1 medication has made that problem much more common. The people on it have real constraints and no practical way to apply them at a table, mid-conversation, with a menu in hand and a server waiting.

The information exists. It is just held in two places that do not speak to each other: the menu, which names dishes, and nutrition databases, which index ingredients.

A diner points their phone at any restaurant or delivery menu. The app reads it, identifies each dish, and flags the high-protein, low-carb and GLP-1 friendly options in seconds. No account setup, no manual entry, no waiting.

It works on printed menus at a table, on delivery menus on a screen, and on photographs taken at an angle in poor light, because that is how menus are actually encountered.

The mobile application is built in React Native, so one codebase serves iOS and Android, alongside a web application for diners who prefer the browser.

AI models handle the menu itself: reading the text off the image and interpreting each dish. Restaurant menus are photographed at angles, printed in decorative fonts, laid out in columns that do not scan cleanly, and written in names that describe a dish rather than its ingredients. A menu says "Tuscan chicken skillet" and expects a human to know roughly what that contains.

Nutrition data comes from third-party databases, which index ingredients rather than restaurant dishes. The inference layer sits between the two, resolving a dish name into likely ingredients and quantities, then against nutrition records that were never structured for this purpose.

The whole path, scan to interpretation to nutrition lookup, is built to resolve fast enough to feel instant, because someone holding a menu at a table will not wait.

Six months from start to live.

The delivered product runs on iOS, Android and the web, with menu interpretation and nutrition resolution working on real menus in real conditions rather than clean digital ones.