A basic understanding of diseases is essential for communities, yet accessing accurate and comprehensible medical information remains challenging, particularly for non-experts in Vietnam. This paper proposes a Retrieval-Augmented Generation (RAG) system for answering disease-related questions, built on a corpus of 610 medical articles collected from youme.vn. The system leverages Halong Embedding, a Vietnamese text embedding model optimized for RAG and fine-tuned on medical data, which boosts retrieval accuracy from 67.44% to 90.57% (Top@10) at dimension 128, and from 55.69% to 79.23% (Top@3) at dimension 256. Experiments demonstrate strong retrieval effectiveness, achieving Recall@10 of 90.57% and NDCG@10 of 0.76902, with accuracy improving by more than 23% after fine-tuning. The main contribution lies in developing a pioneering Vietnamese medical Q&A system that combines multiple retrieval methods—semantic embedding, keyword search (BM25), and structured reasoning via a medical knowledge graph (KG)—to deliver fast and reliable access to information. The strongest performance was achieved with a hybrid configuration (Hybrid-D) integrating all three methods, reaching Accuracy@10 of 90.11%, Accuracy@1 of 72.70%, MRR@3 of 77.70%, and NDCG@10 of 81.67%, significantly outperforming any single component. Finally, Google Gemini is employed as the large language model to generate natural, context-aware answers. The resulting system demonstrates the feasibility of building an end-to-end Vietnamese medical QA platform that effectively integrates hybrid retrieval with advanced generation, thereby improving healthcare information access in low-resource language settings.