Cognisoft Labs collaborated with a confidential Chinese telecom client to implement a federated learning system for predicting customer churn across distributed data silos. The system integrated both horizontal and vertical federated learning techniques, enabling secure collaboration across different data sources without exposing sensitive information. Modern encryption algorithms, including homomorphic encryption and secure multiparty computation (SMPC), were used to protect data during training and model updates.