Ryan-Ray-Martin/Market-Maker ? reverse-engineered prompt

I want to build this research project so I can run the autohedging and market making experiments from the paper.

Please make the single asset and portfolio versions work cleanly, with a simple way to install and start training right away. The agent should learn from simulated client trade flow, choose when to hedge, and adjust spreads to reduce risk, and I want the training to show live progress with plots for rewards and the dashboard. If I run it on a GPU machine, it should use that properly, and it should also save figures and outputs so I can watch progress remotely.

Make sure the code is easy to run from the two main folders, and keep it consistent with the paper backed setup already in the repo. If anything is outdated, please update it enough that it runs smoothly today, and look up current docs online if you need to.