To address the inefficiencies in manual grading and homogeneous feedback for assignments in the Data Analysis and Mining Technology course, this study constructs an intelligent assessment workflow based on the Coze platform. By designing a "local preprocessing-cloud analysis" hybrid architecture and integrating knowledge base retrieval with large language model prompt engineering, the research realizes end-to-end automation of assignment text parsing, code evaluation, and personalized feedback generation. Empirical results show that the solution reduces the grading time for a single assignment from 120 seconds to 6 seconds (a 95% time reduction), and the automated scores demonstrate a significant correlation with instructor evaluations in a sample of 56 students. The findings provide a cross-disciplinary adaptable paradigm for intelligent assessment in the context of educational digital transformation.