A Framework for Conversational Digital Twins: Integrating Generative AI for Operational Event Simulation

Operational planning for large-scale events involves high uncertainty, heterogeneous data sources, and complex resource allocation decisions. Although digital twins and simulation models can support this process, their adoption is often limited by the technical expertise required to configure scenarios and interpret outputs. This paper introduces a Conversational Digital Twin Framework (CDTF) that integrates generative AI with simulation-based decision support, enabling users to configure operational scenarios through natural language. The framework comprises four layers: data integration, a simulation engine combining machine learning and discrete-event simulation components, a conversational AI interface, and a visualization layer for reports and dashboards. It is validated through a TRL4 prototype for football stadium operational planning at the Gran Canaria Stadium, using historical match records, external contextual sources, and synthetically generated operational variables where detailed stadium data were unavailable. The validation focuses on architectural feasibility, workflow integration, conversational configuration, and operational viability. Results show that the agent recognized user intent in 93.3% of evaluated prompts, extracted all explicitly stated parameters in the tested cases, detected all evaluated invalid inputs, and completed the full simulation workflow in all tested scenarios. Average conversational responses took less than six seconds, while complete simulation workflows required approximately 1 min. These findings suggest that conversational interaction can facilitate the access to simulation-based planning and support interactive what-if analysis for event operations, while future work must validate predictive performance with real operational datasets.