Abstract: This study presents a crowd simulation system developed with Unreal Engine to evaluate close contact risks and optimize evacuation strategies in public buildings during health emergencies. Using an agent-based modeling (ABM) approach, the system integrates three modules: input, simulation, and evaluation. It enables quick strategy embedding via parameter mapping, automated model processing through scripting, and real-time recording of close contact events, and output space close contact records and event. A teaching building was used as a case study, testing four scenarios: baseline, staggered door opening, organized guidance, and combined guidance with door-control intervention. Results showed a reduction in close contact events from 9, 456 to 5, 592-a 40.86% decrease. High-risk zones shifted from concentrated clusters to more balanced spatial distributions, with the combined strategy proving most effective. The system supports cross-scenario application, providing a flexible tool for risk quantification in various spatial environments. It offers decision-making support during early design phases and establishes a foundation for future integration of real-world data calibration and multi-agent learning–based optimization.