Enhancing LLMs for Medical Use in Low-resource Settings

Overview

The development of Large Language Models (LLMs) have become increasingly significant in the field of Natural Language Processing (NLP) in recent years due to its remarkable use in instruction understanding and human-like response generation. They have proven to be highly utilised in both academic and practical applications in healthcare. As the technology behind LLMs continues to evolve, their impact on society and industry is potential in further development, making them an increasingly important area of research and development.

This collaborative initiative, undertaken in partnership with the Oxford University Clinical Research Unit (OUCRU), is focused on the development of a Large Language Model (LLM) machine learning system. The primary objective of this project is facilitate the development of a web portal incorporating a LLM designed to engage in contextual discussions concerning Dengue-related topics and other general medical issues.

To achieve this, the proposed LLM will undergo fine-tuning using a diverse array of medical data sources. The aim is to equip it with the capability to offer pre-diagnostic information and provide feedback to users in a manner that closely emulates human-like interaction. The efficacy of the model will be evaluated based on its ability to proficiently handle question-answering tasks while ensuring comprehensibility. The outcomes have demonstrated a notable success, as the fine-tuned models have outperformed the base model across a range of key evaluation metrics. Furthermore, as part of an effort to enhance user accessibility and engagement, this project includes the development of a dedicated website portal featuring a chat interface. This interface will seamlessly integrate with the proposed LLM model, resulting in an improved and user-friendly experience for individuals seeking Dengue-related medical information.

DICA Poster

DICA Methodology

The Cloud Architecture

The cloud architecture for the DICA project is designed to provide a scalable and reliable platform for hosting the LLM model and web chat interface. The architecture is built on Amazon Web Services (AWS) and includes a number of key components, such as Amazon EC2 instances, Amazon S3 storage, and Amazon RDS databases. These components work together to ensure that the LLM model and web chat interface are highly available and performant. The architecture also includes a number of security features, such as VPCs, security groups, and IAM roles, to protect the system from unauthorized access and ensure data privacy and integrity.

Web Chat Interface

The web chat interface is designed to provide users with a seamless and intuitive experience when interacting with the LLM model. The interface is user-friendly and features a clean and modern design that is easy to navigate. Users can input their questions and receive responses from the LLM model in real-time. The interface also includes a number of additional features, such as a search bar, a settings menu, and a news feed. These features are designed to enhance the user experience and provide users with access to a wide range of information related to Dengue fever and other medical topics.