Abstract:In response to the insufficient integration of real industrial scenarios, excessive reliance on "black-box" API calls, and idealized datasets in current large language model courses, a practical case library for large language model courses based on an intelligent power question-answering system is proposed and developed. Guided by the concept of industry-education integration, the case library is developed based on real business data from the Power Safety Regulations, with a full-process practical teaching system covering data governance, knowledge modeling, model optimization, and system application being established through GDR human-in-the-loop data cleaning, vertical-domain knowledge graph construction, parameter-efficient fine-tuning, and question-answering system deployment. Teaching practice and questionnaire survey results demonstrate that students' understanding of the underlying technologies of large language models is deepened, their capabilities in data governance, model optimization, and system development are improved, and their systematic engineering thinking for solving complex industrial problems is enhanced. Practical reference is thus provided for the cultivation of application-oriented large language model professionals in the context of Emerging Engineering Education.