Abstract: To address the pain points in the evaluation of electronic information experimental reports, such as complex data processing, time-consuming assessment by teachers, and susceptibility to biases, this study constructs a DeepSeek + RAG intelligent evaluation system based on a dataset of experimental reports from multiple classes. Through private deployment, it overcomes data volume limitations, and combines knowledge graphs to achieve multi-dimensional correlation analysis. Experimental results show that the system can accurately analyze report contents, generate evaluation results and visualized score distributions, and improve evaluation efficiency. It provides specific technical pathways for the intelligent evaluation of electronic information experimental reports, realizes the “intelligent evaluation-precise feedback-autonomous improvement” teaching closed-loop, and has practical guiding value.