Validation Peak Then Drift

Training loss keeps improving every epoch, but validation Sharpe peaks around epoch 11 and then gradually drifts lower. You are not changing architecture or dataset. What regularization move is most justified?