LEGION develops an AI-driven method for the automated classification and analysis of Roman everyday pottery from the UNESCO World Heritage site of Carnuntum. It builds on a unique dataset of around 70,000 2D vessel profile drawings, enriched with detailed archaeological attribute data (e.g., shape, production, decoration, dating, fragment type) and petrographic thin-section analyses. By combining state-of-the-art machine learning with Explainable AI and human-in-the-loop approaches, LEGION creates a scalable, transparent typochronology that enables reproducible analysis and new insights into production and distribution systems.

The project deliberately relies on established 2D documentation standards rather than complex and costly 3D methods, making the approach practical for day-to-day archaeological work, including commercial archaeology. Beyond classification, LEGION supports the reconstruction of settlement dynamics and wider socio-economic and demographic processes by tracing the distribution of pottery types—revealing phases of growth, crisis, and decline in Carnuntum and shedding light on trade networks, production centres, and cultural identity in the Roman province. All data, ML models, and tools will be made openly available in line with the FAIR and CARE principles and integrated into the European research infrastructure E-RIHS, strengthening Austria’s visibility in Heritage Science and providing reusable digital infrastructure for other regions and periods.