In this talk, we present the randomized Riemannian submanifold subgradient method (RSSM), a lightweight "block-coordinate"-type algorithm for weakly convex optimization over the Stiefel manifold. We show that RSSM finds an $\epsilon$-nearly stationary point in $O(\epsilon^{-4})$ iterations. To the best of our knowledge, this is the first convergence guarantee of a coordinate-type algorithm for tackling non-convex non-smooth optimization over the Stiefel manifold. This is joint work with Andy Yat-Ming Cheung, Jinxin Wang, and Man-Chung Yue.