ETD

• ai
• ai act
• evaluator
• fria
• generalized
• human resources
• iterative design
• modularity
• rights
• user experience
• user interface
• user-centered design
• valuation

The embedding of artificial intelligence (AI) systems in the chain of decisions that influence fundamental rights like access to work or privacy protection makes it necessary to design instruments that can evaluate their effects. This thesis puts forward a generalized evaluator to appraise the risk of discrimination and privacy violation by translating the legal principles into measurable and comparable observables. The system has been deployed as a modular and customizable framework, on the basis of a central orchestrator, autonomous modules, and a a configuration system that guarantees replicability, traceability, and extensibility. It combines the fairness and privacy measures broadly applicable in the literature, increasing the transparency and the accountability of the analyses. A remarkable element of the design is the user interface, which has been transformed from a technical prototype into a friendly and guiding flow through a user-centered, iterative design process. The output appears on interactive dashboards and PDF reporting. Finally, the evaluator’s reliability was verified by reproducing fairness and privacy results from prior work within the new framework and by testing it on two different datasets, thereby confirming both the accuracy of its calculations and its robustness across heterogeneous scenarios, in line with accountability requirements under the European regulatory framework (e.g., the AI Act). The thesis is situated at the crossroads of law, data science, and design, and contributes to building practical tools for the more equitable and transparent use of AI.