November has proven to be a month of outstanding continuous success for E-Group’s research team, marking a significant “winning streak” in the field of medical AI innovation. We are proud to report back-to-back first-place victories at two prestigious scientific events, reinforcing our commitment to academic excellence and technological breakthroughs.
First Success: Best Scientific Article at Neumann Colloquium
The series of successes began on November 14th at the Neumann Colloquium, organized by the John von Neumann Computer Society (NJSZT). We are honored to announce the team comprising Benedek Ágota, Csaba Potyók, and Tamás Mészáros Ph.D won the award for Best Scientific Article. The paper was presented jointly by Ágota Benedek and Csaba Potyók, setting the stage for an incredible week of achievements.
Continuing the Streak: First Prize at BME TDK 2025
Just days later, on November 19, the momentum continued at the Budapest University of Technology and Economics (BME) 2025 Scientific Students’ Associations Conference (TDK). E-Group has been a proud supporter of this prestigious event for years. To demonstrate our deep commitment to fostering the next generation of technological talent, our CEO, Antal Kuthy, personally attended the ceremony.
At this event, Ágota Benedek (co-authored with Csaba Potyók) was awarded First Prize in the Medical Applications category at the Faculty of Electrical Engineering and Informatics competition.
Their winning paper, ” From Guidelines to Graphs: Towards Graph Retrieval and Diagnostic Pathway Augmented Language Models,” represents a significant achievement for our team. Beyond university studies, Ágota works as a Research Engineer at E-Group, actively contributing to our solution development. The work was supervised by Tamás Mészáros Ph.D. (Associate Professor at BME and Senior Researcher at E-Group), whose dual role bridges the gap between academic research and industrial application.
About the Research:
GRaDPaLM While Large Language Models (LLMs) are increasingly used in medical diagnostics, they often lack the structured decision-making capabilities required for safe, clinical use. To address this, the paper introduces GRaDPaLM (Graph Retrieval and Diagnostic Pathway augmented Language Model).
This innovative method overcomes the limitations of standard LLMs by combining static medical knowledge (via Knowledge Graphs) with dynamic, rule-based logic (Diagnostic Algorithms). By integrating these elements, the system significantly reduces hallucinations and ensures that AI-generated recommendations are explainable and grounded in verified medical data.
“The GRaDPaLM system enables the AI to follow standardized decision paths defined by the Guideline Interchange Format (GLIF), allowing for step-by-step, transparent reasoning. Simultaneously, it utilizes a Knowledge Graph Foundation Model (KGFM) to retrieve information from vast, heterogeneous medical sources,” says Ágota Benedek.
The team demonstrated the efficacy of this approach through a prototype focused on dementia diagnosis, validating the system against standard benchmarks.
This success was a collaborative effort. We extend our special thanks to Zita Varga M.D. and Péter Novinszky M.D., who acted as medical experts, ensuring the winning paper was grounded in rigorous clinical reality. We also thank Tamás Mészáros Ph.D. for his expert scientific guidance.
You can access the full TDK paper here: Download Paper
As a science and idea-driven software technology and digital knowledge manufacturing corporation, E-Group takes great pride in its deep academic roots. We are dedicated to supporting talented students and bridging the gap between theoretical research and market-ready innovation.
Upcoming Presentation
Following this double success, Ágota Benedek will present at the upcoming INTERUNI Conference in the section for “Medical Molecular Biomarker Research and AI-based Diagnostics and Therapy Development.”
- Presentation Title: Implementation of a Knowledge- and Protocol-based Medical Diagnostic Assistant.
Congratulations to Ágota Benedek and the entire research team for this outstanding achievement!