Federated learning lets participating devices contribute to a shared model without sending their raw training data to a central server. Explore the architecture, engineering trade-offs and privacy techniques that determine whether this distributed approach is practical.