There are several drawbacks in the assessment of classroom teaching quality in universities, such as limited data sources, long evaluation cycles, and insufficient diagnostic depth, which make it difficult to provide effective support for teaching improvement decisions. This paper proposes a quality education intelligent evaluation architecture based on multi-source data fusion. Through the complete process of data collection - intelligent analysis - closed-loop improvement, the intelligent evaluation of teaching quality is completed. Through research, it was found that the traditional evaluation method is to rely solely on subjective scores at the end of the semester, lacking objective evidence during the teaching process. By leveraging multi-source data fusion, various types of information such as behavior logs, video semantics, academic performance, and supervision evaluations can be collected, thereby achieving precise description of teaching quality and dynamic detection. Therefore, this paper presents a closed-loop mechanism of "evaluation - intervention - improvement", and explores the guarantee means for teaching improvement from the three levels of system, technology, and culture, hoping to provide some references for the digital transformation of the university's teaching quality improvement system.