I would use React for my frontend that does API calls to a FastAPI backend. I wouldn't use FLASK because it doesn't support asynchronous API calls. This FastAPI would be connected to the Tensorflow model of the client within the federated learning network.

You could use docker containers for the different clients and server.