<p>Rocking and torsional ground motions, collectively called rotational, can significantly alter structural response by exciting rocking and torsional modes that are not typically activated by translational shaking. Despite their acknowledged role in amplifying structural damage, current seismic design codes and performance assessment practices do not explicitly account for rotational components. This is particularly due to the lack of recorded ground motions. In this study, rotational ground motions are systematically simulated from translational components of the PEER NGA-WEST2 database using a single-station procedure, following rigorous validation against recorded data from the Kefalonia region in Greece. A global Ground Motion Model (GMM) for rotational acceleration is developed using a feedforward neural network. The model is subsequently calibrated for the tectonically complex Himalayan region through a transfer learning framework to ensure regional applicability. Probabilistic Seismic Hazard Assessment is conducted using the proposed model to develop a region-specific rotational design spectrum for the Himalayas. The shape of the design rocking and torsional components is characterized using the 95th percentile of the normalized Uniform Hazard Spectrum (UHS) at a 2475-year return period obtained from several branches of the logic tree, and peak ground rotational accelerations are presented in the form of a hazard map.</p>