Focusing on the fully English course of Soil Mechanics, this study explores an AI-empowered smart teaching model to address challenges inherent in traditional teaching, such as barriers to understanding English terminology, unclear problem-solving logic, and a lack of personalized tutoring. Relying on the Rain Classroom smart workbench, the course team constructed an AI teaching platform specifically for Soil Mechanics. This platform comprises a structured knowledge base, a reticular knowledge graph, and an exclusive prompt library, enabling the dual functions of a "24H Intelligent Study Companion" and a "Classroom Teaching Companion." By integrating core course resources—including textbooks, exercises, and laboratory guides—the platform supports on-demand questioning and personalized feedback for students while effectively mitigating the phenomenon of model "hallucinations." In terms of implementation strategy, the research adheres to the principles of being "student-centered, data-driven, and technology-empowered." It deeply integrates Large Language Models (LLMs) into the entire process of pre-class preparation, in-class hybrid interaction, and post-class review, thereby establishing a closed-loop smart teaching ecosystem. Evaluation of the teaching practice indicates that this model not only improved students' academic performance but also effectively strengthened their autonomous learning abilities and complex engineering problem-solving skills within an English-speaking context. Simultaneously, the platform's data feedback mechanism optimized teaching strategies and effectively reduced the burden of Q&A on teachers. This study provides a viable smart path for reforming the teaching models of fully English-taught courses in the field of civil engineering.