Quantitative analysis of human motion is essential in healthcare and sports science. Among key biomechanical metrics, 3D body reconstruction and Ground Reaction Force (GRF) estimation play critical roles in understanding movement patterns, detecting abnormalities, and preventing injuries. However, conventional motion capture systems and force plates are expensive and unsuitable for home-based use. In this paper, we propose a novel deep learning framework to jointly predict 3D human mesh shape and foot-ground interaction force from 2D joint sequences. The model leverages a pre-trained 2D pose encoder to extract generalizable representations from detected 2D joint coordinates, which are then decoded by task-specific mesh and GRF heads. The use of a shared, pre-trained 2D pose encoder enables efficient knowledge transfer. Experimental results demonstrate the effectiveness of the proposed method on public datasets.