satyambro03/Traffic-Signal-RL ? reverse-engineered prompt

Build me a small Python demo app around reinforcement learning environments that show three simple cases, email sorting, traffic light control at one intersection, and traffic flow across multiple intersections. I want it to follow the usual environment pattern with reset, step, and state so it feels compatible with OpenEnv style tools.

Make the environments easy to run locally, and include a basic agent that can do inference with repeatable results, even if it just takes simple or random actions as a baseline. I also want a lightweight web demo so someone can open it in the browser and try the tasks, see the current state, choose actions, and view the reward and whether traffic or sorting improved.

Please make it ready to run in Docker and suitable for a Hugging Face Space style demo. Keep the rewards and action choices aligned with the three tasks, and include clear setup instructions so I can launch it quickly. If anything is unclear, look up the current docs online and fill in the gaps sensibly.