PWhiddy/PokemonRedExperiments — reverse-engineered prompt

Build me a Python project that can learn to play Pokemon Red with reinforcement learning, and also lets me try a pretrained agent interactively in the emulator. I want the newer V2 style training flow if that is the recommended one, with a simple way to start training, resume if needed, and run a pretrained model while I can still control the game with keyboard inputs and pause the agent.

Please make setup straightforward on Linux or Mac, and keep Windows notes if they matter. Assume I will provide my own legally obtained Pokemon Red ROM in the project root as PokemonRed.gb. I also want training progress saved as images and viewable in a browser with local metrics, plus the optional live map broadcast feature so a training run can stream to a shared world map.

Use the repo structure and README as the source of truth, keep the original behavior where it makes sense, and wire up anything missing so it actually runs end to end. If anything is unclear, check the current docs online.