Biomedical Chat Assistant with Personalized Document Reader Using BioMistral and RAG

The integration of artificial intelligence in the healthcare sector has revolutionized the way in which medical information is accessed and interpreted. This paper reports the development of a Biomedical Chat Assistant equipped with a personalized reader for documents, using the capabilities provided by the BioMistral framework and Retrieval-Augmented Generation techniques. It processes biomedical data coming from PDF files without structural indexing based on the precise inquiry of the users, providing them with relevant insight. Utilizing advanced natural language processing methods, such as SentenceTransformer embeddings and the LlamaCpp model, the biomedical chatbot provides an interactive conversational interface that increases user engagement. It uses a Recursive Character Text Splitter for effective document segmentation and a Chroma vector store for rapid similarity searches to retrieve relevant information and provide coherent responses. This system, with practical application in health report assessment, has huge potential to be an excellent tool for healthcare professionals and researchers, underscoring the need for AI-driven solutions in real-time medical decision support. It also showcases the integration of RAG systems with domain-specific language models and promises greater accessibility and comprehension of complex biomedical literature.