Is real-time multi-agent RL with Transformers possible in UE5?

I’m primarily an AI/ML student with limited UE5 experience, so I’m honestly not sure how much of this is even technically feasible in a game engine context.

I’m exploring a combat arena concept with 50+ AI agents learning tactics in real-time using reinforcement learning and custom Transformer networks (spatial attention for battlefield awareness, action sequence prediction for opponent modeling, team communication for coordination). This is more of a theoretical exploration than a definite implementation plan.

Can UE5’s Learning Agents plugin realistically handle this scale with real-time learning during matches, or would I need extensive custom C++ plugins for PyTorch/TensorFlow integration? As someone coming from the AI side, I’m not sure what’s reasonable to expect from game engine performance.

Has anyone attempted similar multi-agent RL systems in UE5, or should I be looking at other platforms entirely?