fedpilot/core ? reverse-engineered prompt

Build me a simple federated learning platform called FedPilot that lets me train a model across multiple clients without moving the raw data around. I want a clean web based interface for starting a training run, watching progress, and seeing basic results like accuracy, loss, and client status as it runs. It should work in a distributed way, since this project is meant to use Ray for coordination, and I should be able to run it locally with Docker or a simple command from the README. Please set it up so there’s an easy quick start path, sensible defaults, and a sample workflow I can test right away. If there are config files or environment variables needed, include them in a straightforward way. Also make it easy to monitor what the system is doing, and keep the code organized so I can extend it later for new training jobs or client setups. If you need to check current docs online for anything, go ahead.