POKER-X: An AI-enhanced and GPU-accelerated platform for rapid radiation shielding analysis

The increasing demand for rapid radiation-shielding assessment in nuclear engineering motivated the development of the POKER point-kernel code. To improve computational efficiency in large-scale and CAD-based shielding scenarios, the code has been re-engineered for modern GPU architectures, incorporating hardware-accelerated ray tracing. In parallel, a high-fidelity Photon Shielding Spectra Database (PSSD) was constructed to address the limited accuracy, elemental coverage, and restricted dosimetry scope of legacy buildup-factor datasets. Instead of computing only traditional buildup factors for a few dose quantities, a deep neural network was trained on PSSD to directly predict the post-shielding multi-group photon spectrum. This provides a more general formulation of the buildup process, since any dosimetry quantity can subsequently be derived from the predicted spectrum. POKER-X is therefore presented as an integrated platform that combines a GPU-accelerated point-kernel solver, the AI-based spectral prediction model, and an interactive Qt-based visualization environment. In representative shielding benchmarks, the GPU engine achieved speedups of over 200 × compared with a multi-core CPU implementation, while maintaining dose-rate deviations within 16%–30% relative to Monte Carlo reference solutions. These results demonstrate that POKER-X supports efficient engineering-scale shielding analysis while retaining numerical consistency with established transport methods.