So I implemented Parallel Split Shadow Maps. This works slightly better than Cascaded Shadow Maps. I think that the difference in PSSM is that there is a "crop matrix" which basically "zooms" the split according to the frustum size, and there is only one projection matrix used for the lights POV. The way I implemented my CSM was just rendering the scene with a separate light Projection Matrix for each split. The PSSM way seemed to have completely eliminated the artefacts without any tweaking!
Anyways, this is where I got my implementation details.Oh and here's another movie :)