Running Reinforcement Algorithms on the EVM

Hi guys, I am currently working on an NFT project where we generate a GLTF file to be auctioned daily using reinforcement learning (Machine Learning Algorithm) to generate the art work. I am pretty new the solidity and blockchain tech, and I have heard generating random numbers is particularly difficlut on blockchains. The problem is that the machine learning algorithm I am using to generate the 3D art work (Reinforcement learning) makes random decisions a few thousand times . Is the on-chain approach feasible or do I need to do the art generation off-chain. I would appreciate any help or insight on this problem.

Thank you
Z