Treffer: RAG Chatbot for Healthcare related promptsusing Amazon Bedrock.

Applications of natural language processing (NLP) for use in large language models (LLMs) continue to evolve with technological advancements in the domain Generative AI (GenAI). The massive explosion of data, availability of scalable computing capacity and machine learning innovation, LLMs, have all led towards Generative AI (GenAI) becoming increasingly popular. A major challenge involved with base model LLMs is their tendency to hallucinate. Hallucination in LLMs refers to the output of inconsistent incoherent and sometimes incorrect information or response. This occurs as most LLMs are trained on a large amount of generic data and must be augmented using domain specific and external data for use in GenAI tasks such as chatbots, Q&A, summarization and for text generation. To address the challenge of hallucination, this study will make use of domain specific healthcare data, in the form of PDF files, alongside an FM to create a Retrieval Augmented Generation (RAG) chatbot. This study makes use of the base foundation model, Llama 2 from Amazon bedrock. Our domain specific healthcare data was sourced from relevant and reliable sources. The RAG chatbot was developed using Python and colab notebook and responses were evaluated using Rouge and Meteor, evaluation metrics for automatically generated text. The evaluation was based on three scenarios: responses less than 250 characters, more than 250 characters and combined responses from multiple LLMs. Our findings provide strong evidence that augmenting Foundation models (FMs) with domain specific data can improve the quality of the models’ responses in providing reliable medical knowledge to patients. [ABSTRACT FROM AUTHOR]

Copyright of Journal of Information Systems Applied Research is the property of Information Systems & Computing Academic Professionals (ISCAP) and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)