China has over 150 million diabetes patients, and up to 40% may develop diabetic kidney disease (DKD), the leading cause of chronic kidney disease that can progress to uremia. Western medicine mainly focuses on glucose control, with limited effect on kidney repair, while Traditional Chinese Medicine (TCM) offers unique advantages in early intervention by harmonizing organ function and dredging meridians. This project develops an intelligent TCM decision support system for DKD, consisting of three modules: a multi-level knowledge graph, a reasoning model, and a treatment recommendation model. Multi-source and multi-omics data are standardized and integrated through ensemble learning, while graph neural networks capture complex relationships. A Seq2Seq model generates treatment suggestions, refined by the knowledge graph to ensure consistency and precision, providing reliable support for DKD prevention and management.