tzz1289328163-tzz/one-sentence-ai-food-ordering

One Sentence AI Food Ordering is a prototype that turns a single natural language request, such as I want something light, under 30 yuan, and delivered fast, into a food delivery recommendation. It understands taste preferences, budget, delivery time, dietary restrictions, and address from one sentence, then suggests one to three matching restaurant options. The project is built as an explainable agent. Instead of just showing a recommendation, it displays the underlying intent it detected, the entities it extracted from the sentence, the missing information it still needs, its confidence level, any clarifying questions it asks the user, and the current step in its process. The full flow moves from a chat style request to intent recognition, entity extraction, filling in missing details, deciding whether to ask a clarifying question, searching for matching restaurants and dishes, filtering and ranking results, showing recommendation cards, adding items to a cart, and confirming the order. The application covers the complete ordering path, from the initial chat recommendation to a restaurant detail page, a shopping cart, order confirmation, and order history. The frontend is built with plain HTML, CSS, and JavaScript, and runs as static pages. The backend is a small Node.js and Express server that proxies requests to a GLM AI model from Zhipu for the natural language understanding and streaming responses. Everything can run locally without complex build tooling. To try it, a user copies the example environment file, adds their own Zhipu API key, installs the server dependencies, and starts the Express server, then opens the chat page in a browser. Restaurant and dish data is seeded locally rather than pulled from a real backend service or database. The README lists several directions for future work: connecting to a real map service for delivery addresses and distance based delivery fees, adding a review step before payment for address, amount, and coupon changes, building user profiles from order history, supporting multiple AI models instead of just GLM, and replacing local browser storage with a real database so actual restaurants could use it.