In this paper, we present our system for the CCKS2024-TCMBench. TCMBench is a comprehensive benchmark designed to evaluate the performance of large language models (LLMs) in the traditional Chinese medicine (TCM) domain. Specifically, TCMBench defines two tasks: (1) TCM question answering with explanations, and (2) TCM natural language inference. For Task 1, we first supplemented the knowledge points associated with questions in the CMExam dataset. During the question-answering process, we retrieved relevant knowledge points by matching similar historical questions, thereby enhancing the LLMs’ answering performance. For Task 2, we calculated sentence similarity between premises and hypotheses, determining the levels of similarity based on the computed values, which served as supplementary input to enhance the LLMs’ reasoning capabilities. By applying these methods, our system achieved second place in the CCKS2024-TCMBench (Non-Finetuning Track), validating the effectiveness of the proposed approach.