Objective To develop a system for retrieving information from clinical records of Traditional Chinese Medicine(TCM)based on generative large language models(LLMs).Methods Applicational needs of the system were analyzed,and entity types to be retrieved were identified.The functions,workflows,and architecture of the system were designed by combining the automatic retrieval capabilities of LLMs with human-in-the-loop(HITL).The software was developed using such frameworks as vLLM and Node.js.Interaction of multiple commercial/open source LLMs was implemented using OpenAI-compatible interfaces.The quality of information retrieved from LLMs was enhanced by prompt engineering.Results This system supported task allocation,automatic retrieval of structured information,and manual review.To evaluate its performance,the moonshot-v1-8k model was used to retrieve clinical records of TCM before manual edition was performed.Combining large language model pre-annotation with meticulous annotator edits improved accuracy by 26.6％compared to the BERT-BiLSTM-CRF model,and enhanced extraction efficiency by 1.6-fold relative to purely manual methods.Conclusion General generative LLMs can retrieve a wide range of entity information from TCM records with high accuracy and scalability.The design and implementation of this system approach may provide a useful reference for developing other biomedical information retrieval systems.