I don't want to give you bad news, but there's no way to put it in a different way.
ndnSIM last update was 2 years ago (a very long time), and the ns-3 sources did evolve a lot in the meantime. Not only we're using cmake, we also moved away from pybindgen, so the python interface is quite different in some very important points.
The problem is that as far as I know ns3-ai uses the "new" python interface, so it's not just a matter of cmake.
One possibility could be to adapt ns3-ai to ndnSIM, but I have no idea if this is possible, or how hard it could be.
Another possibility could be to "revive" ndnSIM and forward port it to use a newer ns-3 version as its base, but even this might be a very complex task, as ndnSIM made some "fixes" to its own ns-3 base system, and I don't know which ones have been accepted to the ns-3 code, and how many breaking changes we made in the meantime.
The third alternative is to just use ndnSIM as a guideline and reimplement it from scratch. Sometimes it's easier than forward porting.
So, overall, I'd say that the alternatives go from "ouch" to "triple ouch", and I don't know which one of them is the less painful one. For sure there's none classified as "easy".
I also checked if a different simulator exists. It does: NDNOMNeT, last updated... 4 years ago. Doesn't look good either.
Sorry...