The project HealthLight aimed at addressing Vietnam’s need for accessible and understandable public health literature using the advanced Retrieval Augmented Generation (RAG) architecture.
Contributors
Anh Truong, Minh Nguyen, Thach Ho, Phuoc Dinh
Program
Bachelor of Engineering (Software Engineering) (Honours)
Notable achievement
Best Software Engineering and IT Project in the School of Science, Engineering & Technology capstone projects showcase
Bridging health information gap through Al
In Vietnam, the lack of accessible public health literature hinders the management of infectious diseases and chronic conditions, leading to complications and deaths. Hence, the project HealthLight aimed at addressing Vietnam’s need for accessible and understandable public health literature. HealthLight specifically proposes the development of a medical Question-and-Answer (QA) system utilizing the advanced Retrieval Augmented Generation (RAG) architecture. The HealthLight project will contribute to understanding the practical implementation of advanced RAG and modular RAG systems in a medical setting, particularly the use of a Knowledge Graph (KG) index to retrieve more precise information.
Beyond the research purpose, the project aims to empower individuals to manage their health proactively by providing clear and understandable medical information based on the latest research. This will improve public health literacy in Vietnam, aiding in better disease prevention, early diagnosis and informed healthcare decision-making for both the general population and healthcare professionals.
Methodologies and architecture
HealthLight full working flow
RAG (Retrieval Augmented Generation) mitigates hallucinations by retrieving factual data from large datasets, ensuring accurate responses. It also streamlines complex workflows, integrates with other tools and keeps the LLM up-to-date for reliable information generation.
Results
HealthLight system components:
- The Retrieval module
- The medical text generation module based on retrieved data
- The knowledge graph integration module for enhancing accuracy and contextualization of generated content
This structure helps reduce the "language hallucination" phenomenon commonly encountered in text generation models, ensuring that the information remains up-to-date and is clearly grounded in reputable medical documents. The system also supports users by tracking interaction history, enabling follow-up questions, and enhancing the understanding of complex medical issues through a user-friendly interface.
Conclusion
HealthLight showcases the integration of advanced Retrieval-Augmented Generation (RAG) and knowledge graphs to provide precise, in-depth answers to medical questions. The system facilitates access to the latest research and empowers users to engage in follow-up inquiries, enhancing their understanding of complex medical information through a user-friendly web application.
