Péter's winning entry, "A Composable Agentic System for Automated Visual Data Reporting", stands out for its focus on teamwork between humans and AI. Instead of a single, opaque AI doing everything, his system breaks the process into transparent, modular steps: it cleans and profiles data, spots key insights using statistical tools, designs charts based on formalized visualization design best-practices, and outputs two user-friendly formats: an interactive web report for exploring data hands-on, and editable notebooks for digging deeper into the "why" behind each finding. This multi-agentic approach makes the AI reliable, fixable, and collaborative, avoiding the pitfalls of brittle, all-in-one models.
You can explore the project, including example reports and code, at peter-gy.github.io/VISxGenAI-2025. The work was done in collaboration with Dominik Moritz of Carnegie Mellon University, Torsten Möller of the University of Vienna and Laura Koesten at Mohamed bin Zayed University of Artificial Intelligence in Abu Dhabi. Congratulations, Péter—well deserved!
