This study introduces a tutoring system for college physics experiment courses powered by generative artificial intelligence technology, with statistical analysis on the dialog collected from pilot classes. The system integrates the DeepSeek-V3 large language model with a dynamic knowledge base, enhancing adaptability of the general model in educational settings. Through dual-role interaction design and answer tracing mechanisms, it inspires students’ interest while improving the standardization of response content. The backend database archives users’ dialogues, enabling unobtrusive research of learning process of students. This platform facilitates personalized and differentiated instruction, offering a viable solution to reconcile the tension between sizable student population and personalization in general physics laboratory courses. The dialogue corpus collected through prolonged observation further provides data support for teaching evaluation and reform.