面向对象与背景
YiSu Hotel was designed for modern travel scenarios as an integrated hotel booking service. It combines a mobile booking experience, a merchant and administrator console, and a conversational AI recommendation workflow in one project.
需要解决的问题
Hotel discovery and booking often require users to repeatedly combine city, date, budget, rating, and facility filters. At the same time, hotel operators need a practical workflow for maintaining rooms and hotel information, while administrators need review and publishing controls.
项目目标
- Provide a responsive mobile journey covering hotel search, filtering, details, room selection, and booking.
- Build a PC management system for merchant hotel maintenance and administrator review and publishing.
- Support Chinese, English, Japanese, and Spanish interfaces.
- Enable natural-language hotel discovery through a RAG recommendation agent.
设计约束
- The project uses JSON files as development storage rather than a production database.
- AI recommendations depend on external embedding and language-model services.
- The three services run independently and require coordinated API and CORS configuration.
- The project is retained as a completed demo and is no longer actively maintained.
搭建的逻辑框架
- Separated the system into an Express API server, a React and Vite mobile client, and a React and Vite administration client.
- Implemented JWT authentication, merchant and administrator roles, hotel and room CRUD, image uploads, and review and publishing workflows.
- Built the recommendation agent with Qwen text embeddings, cosine-similarity retrieval, intent extraction, Top-K matching, and Moonshot-generated responses.
- Added conversational history and recommendation cards, plus four-language switching through i18next.
使用价值
- Completed an end-to-end demo spanning customer booking, hotel operations, administrative review, and AI-assisted discovery.
- Supported natural-language filters such as city and budget and returned clickable hotel recommendation cards.
- Created a reusable three-part architecture with documented APIs and local startup instructions.
项目总结
- A useful recommendation agent needs structured retrieval and filter extraction in addition to fluent generation.
- Role boundaries, content lifecycle, and data contracts need to be designed together across customer and administration products.
- The next production iteration would replace JSON storage, harden account security, and automate vector-index updates.