Nowadays AIs tend to use TensorFlow.
module can't create new models. It can only load pre-trained models. (Well, it probably can if you know all the deepest magic of Tensorflow, C, MX and Libtensorflow.pm. I don't)
a perfectly good solution until cuda 12 broke the entire build- and tool chains and Apache retired the MXNet project.
for a Dockerfile that builds and runs AI::MXNet and get Perl to create your next AI!