Transforming Healthcare with Gen AI: An Innovative Biomedical Chatbot Leveraging Mixture of Experts and Intelligent Retrieval Systems

This paper presents the development and evaluation of an advanced biomedical chatbot designed to enhance patient question-answering capabilities. The chatbot utilizes a mixture of expert model architecture and a retrieval-augmented generation (RAG) system. One LLaMA3 model is fine-tuned on a dataset of patient questions and doctor answers dataset. In contrast, another is fine-tuned on a broader biomedical question-answer dataset, including summaries of doctor-patient interactions and related texts. These models are quantized and combined via a router to form a robust mixture of expert systems. Additionally, a vector database is created to catalogue medications for various ailments and symptoms, which the RAG system uses to retrieve relevant information during patient interactions. The integrated system is deployed as an interactive chatbot that provides accurate and contextually relevant responses to patient inquiries. The chatbot’s performance is rigorously evaluated using multiple biomedical datasets, demonstrating its potential to improve patient engagement and healthcare delivery by offering reliable and timely information.