MedFusionQA: A Retrieval-Augmented Large- Language Model for Biomedicine

In recent years, transformer-based language models such as BERT and its biomedical variants like BioBERT and GatorTron have demonstrated exceptional performance in various natural language understanding and information retrieval tasks within the medical domain [19], [20], [21]. However, these models often exhibit limitations when deployed in realworld applications, particularly due to their computational inefficiency and challenges in adapting to multilingual contexts [6], [10]. Additionally, general-purpose models may hallucinate factual information or lack the domain specificity needed for accurate medical insights [8], [9]. To overcome these shortcomings, we introduce MedFusionQA, a lightweight, multilingual question-answering framework specifically tailored to extract and deliver relevant information from medical documents. Leveraging DistilBERT for its efficiency and performance trade-offs [19], our system incorporates PyMuPDF for extracting data from clinical PDFs, integrates Google Translate for multilingual question support, and uses a simple FLASK-based backend to enable smooth user interaction. This pipeline offers a low-latency and easily deployable solution for real-time biomedical question-answering. MedFusionQA bridges the gap between domain-agnostic language models and the critical need for precise, accessible, and efficient healthcare information systems, addressing a key gap identified in prior biomedical NLP research [1], [2], [7], [12], [18].