Healthcare Diagnostic RAG-Based Chatbot Triage Enabled by BioMistral-7B

Retrieval-augmented generation (RAG) approaches represent a significant step forward in patient triage and assistance, effectively enhancing user engagement and diagnostic accuracy when combined with the development of chatbots. This paper proposes an advanced design of a chatbot system aimed at enhancing support for the triage and diagnosis of patients through the utilization of sophisticated models like BioMistral7B and PubMedBert. It employs these models along with the Conversational Retrieval Chain (CRC) in effectively handling user queries that fall under core medical and triage-based questions. Accordingly, the core medical reference books–among which are included The Merck Manual of Diagnosis and Therapy, Harrison's Principles of Internal Medicine, The Gale Encyclopedia of Medicine, and The Oxford Handbook of Clinical Medicine–are converted into the form of vector embeddings and stored in the Qdrant vector database for subsequent quick retrievals. The performance evaluations further proved that the chatbot was able to attain responses within the range of 22 to 35 seconds, with an average response time of 28 seconds, when tested on modest hardware setups, while the BLEU & ROUGE scores reflect its outputs to be of decent quality. The findings reveal that AI-driven chatbots hold promising possibilities in enhancing diagnostic accuracy as well as improving patient engagement in various healthcare settings.