This study explores the development and implementation of an AI-based online translation teaching system with a layered architecture, including data, foundation, business, and user layers. The data layer integrates SQL and NoSQL databases for efficient storage and querying of various data types. The foundation layer combines system management with intelligent services, utilizing large models and multimodal technologies to enhance teaching efficiency, support blended learning, and drive the digital transformation of resources. The business layer supports four key platforms---translation teaching, training, self-learning, and testing---along with an intelligent terminal service platform. The user layer, with four interfaces based on participant roles, ensures smooth system operation. A microservices architecture improves stability and resource utilization. Built on the Langchain framework, the intelligent foundation has been tested for reliability and performance, and the terminal service can handle multiple concurrent connections and process students' audio data accurately.