This paper presents GHOST (Gaussian Hand-Object Splatting), a fast category-agnostic framework for reconstructing dynamic hand-object interactions from monocular RGB videos. The method represents both hands and objects as dense, view-consistent Gaussian discs to achieve complete 3D reconstructions. Three key innovations are introduced: a geometric-prior retrieval and consistency loss for completing occluded object regions, grasp-aware alignment for refining hand translations and object scale to ensure realistic contact, and a hand-aware background loss that prevents penalizing hand-occluded object regions. The framework enables physically consistent and animatable reconstructions while running an order of magnitude faster than existing methods, with applications in AR/VR, robotics, and embodied AI.