Third place in Medical Track - The team of Wang Rongsheng

Coming from the Faculty of Applied Sciences of Macao Polytechnic University, our team previously focused on medical imaging and AI research. With the explosion of ChatGPT, we quickly did extensive research exploration based on our previous accumulated medical research experience and big models. Participating in the Baichuan Intelligence AGI Hackathon is the best proof for us to show our research results in the past period. Projects Through continuous and deep cooperation with hospitals, we are constantly exploring the possibility of landing AGI in medical scenarios. In this competition our team proposed CareGPT, a trusted medical big model solution based on Baichuan2-13b-Chat and RAG. Specifically, we use Baichuan2-13b-Chat as a base to assign patients to specific departmental Agents through guided diagnosis (or intent recognition), and the specific Agents diagnose the patient's disease and make scientific suggestions through multi-round dialogue. In order to enhance enhance the interpretability of medical LLM in practice, we introduce a posteriori methods to combine the medical recommendations generated by medical LLM with enhanced RAG (double retrieval with fine and coarse ranking) recall validation, which greatly improves the accuracy and interpretability of medical LLM diagnosis. Besides, we comprehensively evaluate the medical LLM fine-tuned based on mainstream Chinese LLMs from the hallucination rate and CMB medical list in our project, which proves the great possibilities of Baichuan2-13b-Chat in medical landing. We open-source the whole project, including ten million medical training corpus and more than 30 medical macromodels: