<p>Generative Artificial Intelligence (AI), driven by advancements in generative large language models (LLMs), is leading a new wave of technological innovation and emerging as a key factor in global competition. As these models increasingly influence various sectors, they open new opportunities for digitalization and innovation in education. This study leverages the inferential capabilities of the large language model GLM-4 to evaluate pre-service teachers’ teaching practicum in real-time, providing immediate, targeted feedback to address discrepancies in teaching performance and helping pre-service teachers improve their skills. It addresses several challenges in traditional evaluation methods, such as the absence of real-time assessment, the difficulty of offering personalized feedback, and the inefficiencies in conventional pre-service teacher training. A virtual reality (VR) environment creates a simulated classroom, offering pre-service teachers an immersive platform to practice and improve their skills. The framework utilizes an end-to-end speech recognition model based on Paraformer, enabling accurate, real-time evaluations with immediate feedback and constructive suggestions. Furthermore, a comprehensive teaching practicum evaluation system has been developed, incorporating a computational model, optimized prompts, and specialized course knowledge bases to improve the accuracy and objectivity of the evaluations. This study explores the impact and challenges of applying LLMs in course evaluations, reviews the current state of such evaluations, and proposes strategies for continuous optimization and improvement, aiming to advance reforms in pre-service teacher training and support the high-quality development of teacher education.</p>