Health-LLM is a personalized retrieval-augmented disease prediction system that integrates large-scale feature extraction and medical knowledge scoring to improve disease prediction and personalized health management. The system uses a retrieval augmented generation (RAG) mechanism to enhance feature extraction and incorporates a semi-automated feature updating framework to improve prediction accuracy. It leverages the Llama Index framework to score features and uses XGBoost for disease prediction. The system was tested on a large number of health reports and outperformed existing methods, achieving an accuracy of 0.833 and an F1 score of 0.762. The system's key contributions include the integration of LLMs with medical knowledge for personalized disease prediction and the use of RAG for enhanced feature extraction. The system also provides personalized health advice based on predicted health risks. The system was compared with traditional methods and other large language models, demonstrating superior performance in disease prediction. The system's effectiveness was validated through a case study using the IMCS-21 dataset, where it achieved an overall prediction accuracy of 83% for disease. The system's components were analyzed, showing the importance of indexing professional healthcare data and effective data processing with CAAFE. Future work includes integrating multimodal data such as medical images to further improve disease prediction capabilities.Health-LLM is a personalized retrieval-augmented disease prediction system that integrates large-scale feature extraction and medical knowledge scoring to improve disease prediction and personalized health management. The system uses a retrieval augmented generation (RAG) mechanism to enhance feature extraction and incorporates a semi-automated feature updating framework to improve prediction accuracy. It leverages the Llama Index framework to score features and uses XGBoost for disease prediction. The system was tested on a large number of health reports and outperformed existing methods, achieving an accuracy of 0.833 and an F1 score of 0.762. The system's key contributions include the integration of LLMs with medical knowledge for personalized disease prediction and the use of RAG for enhanced feature extraction. The system also provides personalized health advice based on predicted health risks. The system was compared with traditional methods and other large language models, demonstrating superior performance in disease prediction. The system's effectiveness was validated through a case study using the IMCS-21 dataset, where it achieved an overall prediction accuracy of 83% for disease. The system's components were analyzed, showing the importance of indexing professional healthcare data and effective data processing with CAAFE. Future work includes integrating multimodal data such as medical images to further improve disease prediction capabilities.