Deep learning-based Medical Question Answering (MQA) systems are transforming access to healthcare information by enabling accurate and timely responses to complex queries. However, existing systems face challenges such as limited large-scale, high-quality medical datasets, inadequate contextual understanding, and difficulties in managing diverse medical terminologies This research proposes a novel closed-domain MQA system that addresses these limitations through innovative methodologies. The system employs BioBERT-based domain-specific embeddings trained on biomedical literature to accurately capture medical terminology, abbreviations, and contextual nuances. To model sequential dependencies in queries, Recurrent Neural Networks (RNNs) are integrated, enabling contextual interpretation across longer text sequences. Additionally, a question expansion mechanism utilizing medical dictionaries and ontologies like UMLS addresses synonymy, ambiguity, and terminological variations, ensuring that diverse medical expressions map to consistent, semantically relevant concepts for precise answer retrieval. Extensive evaluation using metrics such as F1-score, precision, recall, and exact match demonstrates the system’s superior performance compared to existing models. The key contributions include improved contextual understanding, better handling of medical terminology, and a scalable framework for future medical NLP applications. This system not only offers a reliable tool for healthcare professionals and patients but also advances the field of intelligent question answering by supporting evidence-based clinical decision-making.