This study introduces an approach for constructing a specialized question-answering system for the medical domain through the application of large language models. To improve answer accuracy and reliability, the system adopts the retrieval-augmented generation framework, which enables the integration of external knowledge sources by retrieving relevant information to support response generation. Combined with prompt design and reinforcement learning techniques, the model adapts to updates in medical data. Using a large Vietnamese language model, the system achieves high performance in diagnosing diseases and answering medical-related queries. Experimental results show that the system provides accurate and context-aware answers in the field of healthcare consultation, demonstrating its potential to improve medical chatbot systems. Key techniques include the integration of a multiagent architecture, long-term memory, and a hybrid search method for better information retrieval, ensuring personalized and reliable responses for users.