Federated Learning is a rapidly growing field that enables the decentralized training of AI models while keeping data on the premises where it is generated. This approach enhances data privacy and security, improves scalability and adaptability, reduces communication and bandwidth usage, and ensures compliance with AI and data protection regulations.

After evaluating the open-source frameworks available on the market, it was decided to develop a tool based on the pipes and filters paradigm. In this framework, “Pods” areused as atomic units of processing, and “Wires” to establish the connections between them. This design allows a more easy creation of flexible and customized solutions for the clients, adapt to complex federated learning scenarios or topologies (swarm learning, hierarchical learning), and also implement reconfigurations and retraining if the model's performance in production declines.

Status: submitted to the European Patent Office.