Design and Implementation of Medical Chatbot Using Gamma LLMV2

In this paper, an intelligent medical chatbot ispresented, which can deliver context-related medical informationabout the symptoms, medications, and general health tips. Thesystem consists of a massive language model based on the LLaMAarchitecture of Meta and semantic retrieval involving vectors,meaning medical reference documents are sliced in meaningfultext slots and converted to vectors. The medical reference dataset,GALE Encyclopedia of medicine, a 637 page dataset, is theprimary source of knowledge and is divided into text segments sothat it can be easily searched by semantic queries. Theseembeddings are stored and located through the use of thePinecone vector database which allows one to perform fastsemantic similarity matching when the user is interacting with theplatform. The chatbot is developed as a web-based application onthe Django framework with the assistance of LangChainmanaging the interaction between the retrieval system and thelanguage model, and LlamaCpp enables the language model to behosted locally without the use of external APIs.A retrieval-augmented generation method is used to querydownloads the user query and retrieves the pertinent medicalinformation by accessing the Pinecone vector index, which theuser is then sent and the response is then generated, and the chathistory may also be used to achieve multi-session conversations.Accuracy, the speed of response, and the quality of interactionsbetween the user are used to measure the performance of thesystem. Based on experimental outcomes, the chatbot providesreliable and contextually appropriate medical responses withmoderately good response times on clock speeds with CPU-basedsystems. The paper demonstrates that combining Pinecone-basedvectors retrieval with large language models provides a highly effective and privacy-sensitive approach to providing healthcare-related information, where further enhancements to the system include real-time information updates and multi-languagesupport.