The idea came from watching the same scene play out in restaurants over and over: a guest holding a menu, waiting to ask a busy waiter something basic - "is this spicy?", "does it have gluten?", "what do you recommend?" Meanwhile the owner is reprinting menus every time a price changes. The menu is the most-read document in any restaurant, and it's still a static piece of paper that can't answer a single question. I'd been building RAG (retrieval-augmented) chatbots, and the insight was that a menu is a perfect knowledge base: small, structured, high-value. But the real blocker for restaurant owners isn't the chat - it's data entry. Nobody will type 80 dishes into a dashboard. So the approach flipped: the product had to start from a photo of the existing paper menu. A vision model reads it like OCR - handwritten items, multi-column layouts, chili icons, Thai or any other script - and outputs structured dishes with prices, translations, dietary tags, and spice levels. The owner just reviews and publishes. The process evolved a lot along the way. Early versions let the AI answer freely, and it would occasionally invent dishes - unacceptable when allergies are involved, so answers are now grounded strictly in the restaurant's own menu data, with hard filters for price, dietary needs, and spice. I also learned that "I couldn't find that" is a dead-end answer; the bot now always suggests the closest real alternatives instead. And owner feedback shaped small things that matter, like marking signature dishes so "what's your best seller?" gets the answer the owner actually wants. The goal ended up simple: paper menu in, a menu that speaks every language and answers every question out - in minutes, with zero data entry.

Menu Gen AI

Paper to Digital Menu in ~30 seconds using AI

Paper to Digital Menu in ~30 seconds using AI