We propose FGBoost, a gradient boosting framework designed to intrinsically model complex regression relationships with non-Euclidean outputs in geodesic metric spaces.