Back to Portfolio

Menu AI

Restaurant menu in, finished POS build out. Menu AI turns a restaurant's PDFs, photos, spreadsheets, and online-ordering links into a complete POS menu build that a person reviews and Menu AI pushes straight into the POS.

Role Founder & Lead Engineer
Type B2B SaaS
Stack Python, FastAPI, React, Claude + GPT
Status Live, 270 menus in POS

Three systems, one pipeline

AI Document Reading

AI models read what the restaurant sends (PDFs, photos, spreadsheets, online-ordering links) in whatever layout their menu uses.

AI Engine

The models turn what they read into a complete POS menu build: items, sizes, prices, modifier groups, and kitchen routing.

POS Push

After a person reviews and approves the build, Menu AI pushes it straight into the POS. A menu only counts as live once that push lands with no errors.

What it's built with

Python + FastAPI BACKEND
React 19 + Vite (TypeScript) FRONTEND
Claude (Agent SDK) + OpenAI AI
PostgreSQL DATABASE

Problems worth solving

Messy input, clean output

Restaurant menus arrive as PDFs, photos, spreadsheets, and online-ordering links, in whatever layout the restaurant uses. AI models read all of it and turn it into structured data.

The AI builds the menu, not just a transcript

Reading a menu is the easy part. The output has to be structured the way a POS expects it, with sizes, modifier groups, and kitchen routing all linked, not just a list of items and prices.

A person checks every build

A person signs off on every build before it goes live. The AI does the keying; the technician checks the result. See the production numbers.