Welcome @Ivaylo.IG,
There is always a way, but the question is often, if it’s worth pursing! In my opinion, polysurfaces really are not the right choice here, unless each polysurface represents a discrete part of the final geometry.
For the original, they first did a sub-d model with Rhino and t-splines, which is not available anymore:
This seems absolutely on track with what you did with sub-d modelling in WIP 7 (?).
Now, where your workflows diverge is at the next step. ZAHA use a topological mesh, which they probably produced with an advanced script, in modelling software with manual remeshing, or even both (?):
The great advantage here is that their mesh includes much of the necessary information to proceed further. The mesh edges are used for the facade pattern, the mesh faces and vertices contain adjacency information, which probably comes in hand later, etc.!
The many individual surfaces of your brep currently only try to mimic the initial geometry and don’t inform on much more.
Let’s say you pursue this route! You already seem to have projected some kind of point pattern onto the volume. Now you need a custom script the can compute geodesic curves for subsurfaces of breps!
You could probably start at a point, get the points closest subsurface, get this surfaces adjacent subsurfaces, and check if they have a closest point. If they do, then you would need to project these adjacent points to the closest point on the initial subsurface and connect them geodesic-ally to the initial point. You would end up with geodesic curves fragments, but since your initial information was so weak, this will probably plague you throughout the next steps.
How are points currently structured? How do you produce these?
Here’s the video that I got the screenshots from:
