Federated Learning (FL) enables collaborative model training without centralizing raw data, but distributed deployments remain exposed to typical poisoning and inference attacks and must operate across resource-constrained edge environments. The REMINDER project addresses these challenges by designing an edge-centric framework that provides privacy and security mechanisms with byzantine-robust learning approaches. This paper reports some of the project’s mechanisms and their implications for the development of robust and secure FL deployments, including: (i) a threat model addressing poisoning and inference risks; (ii) a modular architecture with differential privacy, secure authenticated updates, and robust aggregation against malicious clients; and (iii) two representative validation scenarios, such as eHealth and smart buildings, which ground design choices and highlight domain-specific constraints. Building on these contributions, the present work formalizes the end-to-end workflow, specifies component interfaces, and links attack classes to concrete mitigations within REMINDER, while outlining open challenges such as verifiable aggregation and the privacy–utility trade-off introduced by differential privacy in common FL settings.