In a cafe or restaurant the menu changes every season, and the staff even more often. A waiter has to know the composition of forty dishes, the fourteen allergens, which dishes are vegan, what can be swapped to be gluten-free and what happens when a guest asks for lactose-free. Recipe cards with exact composition sit in the costing module of Lightspeed or Poster POS, but only the chef and the bookkeeper see them, the allergen table hangs in the kitchen on paper, and the waiter answers from memory. A mistake here means not an unhappy guest but a health risk and liability for the venue.
We build a ChatGPT Custom GPT with your recipes, allergen tables and substitution rules. We take recipes from the recipe card export of Lightspeed or Poster POS, so that the composition is the same one the kitchen and costing use, and we compile the allergen marking together with the chef, covering the 14 allergens listed in EU Regulation 1169/2011. Staff ask on a phone or tablet at the table, and the assistant answers only from the uploaded documents and names the recipe card or table row each answer came from. If a dish or ingredient is not in the documents, the assistant says so and refers the question to the chef instead of making something up. Guests can be given a separate, simpler version with approved allergen tables via a QR code on the menu. When the chef changes a recipe in the POS, we refresh the assistant's knowledge base from a new export on an agreed schedule and remove old versions. The model runs on the OpenAI platform without training on your data, the recipes remain your property, and every question is visible in the log, so you can review what staff ask most often.