A new ChatMED research contribution was presented as a poster at the 7th International Conference on Deep Learning Theory and Applications (DeLTA 2026) in Porto, Portugal.
The poster, entitled “A Decision Pipeline for RAG Architecture Selection in Medical Education: Empirical Analysis and Cost–Performance Guidelines,” presents a practical framework for selecting an appropriate Retrieval-Augmented Generation (RAG) architecture for medical education applications. The work was authored by Jovana Dobreva, Tadej Horvat, Matjaž Gams, Igor Mishkovski, Kostadin Mishev, and Monika Simjanoska Misheva, representing Ss. Cyril and Methodius University in Skopje, the Jožef Stefan Institute in Ljubljana, and Aalborg University.
The study compares Vanilla RAG, Vector RAG, Graph RAG, and Hybrid RAG using the RAGCare-QA dataset, comprising 420 medical multiple-choice questions across six specialties and three difficulty levels. Rather than finding a clear accuracy winner, the evaluation showed that all four architectures achieved approximately 96% accuracy, with no statistically significant pairwise differences.
This leads to the poster’s central message: when accuracy is comparable, RAG architecture selection should be driven by the intended workload, latency requirements, computational cost, and complexity of the questions. The proposed three-phase decision pipeline therefore combines requirements analysis, constraint evaluation, and architecture assessment to support more evidence-based system design.
The work contributes to ChatMED’s broader efforts to develop efficient, evidence-based and practically deployable AI architectures for healthcare and medical education.
