Adesso Exploitable Results and Business Impact

Within the project, Adesso has developed valuable technical know-how and tangible results that can be exploited both commercially and scientifically. The primary exploitable outcome is a scalable, privacy-preserving recommendation framework based on Federated Learning and cross-domain collaboration, enabling organizations to deploy intelligent recommendation services while maintaining data privacy and supporting domain-specific customization.

A privacy-preserving recommendation framework that combines cross-silo Federated Learning with Two-Tower recommendation models and a server-side explainability layer based on TSK fuzzy rules. The solution enables collaborative model training without sharing raw data, provides global explainability through FED-XAI, supports scalable deployment via NVFLARE, and generates interpretable recommendation insights while preserving user privacy.

The solution requires a federated learning infrastructure with participating organizations capable of local model training and secure model aggregation. Integration depends on schema compatibility between partner environments, secure communication channels, and deployment within privacy-compliant infrastructures. Existing recommendation pipelines can be integrated with limited backend modifications provided that compatible Two-Tower models and federated orchestration are available.

Organizations that develop or operate recommendation systems and require both personalization and strong data privacy guarantees. Target customers include media and streaming platforms, e-commerce companies, online marketplaces, digital content providers, financial service providers, healthcare organizations, and enterprises implementing AI-driven personalization solutions in regulated environments.

The solution can be reused by organizations operating federated learning environments that require privacy-preserving recommendation and explainability capabilities. Reuse is subject to compliance with applicable data protection regulations, compatibility with existing federated infrastructures, and appropriate licensing of the developed software components. The modular architecture allows adaptation to different application domains and recommendation scenarios with minimal customization.