Press 1 for the Future Human-Centred Conversational AI in Public Administration

Public administration stands at a crossroads: citizens expect immediacy and simplicity, yet institutional interactions often remain slow and complex. Conversational AI offers a promising path towards more accessible public services, where natural language bridges people and institutions. This thesis explores how conversational agents can be designed and governed across progressive levels of citizen-service automation, from first-contact support, to guided information retrieval, and finally to personalised assistance based on verified identity. The study is grounded in three projects developed during an internship at Dinova S.r.l. in Bologna: Concilia, a voicebot for traffic fines and appeals; Sportello Telematico, a municipal information chatbot; and SPEED Tributi, a taxation assistant with digital identity authentication. The internship focused on analysing and improving these systems through iterative conversation design, strategic prompt engineering, robust validation logic, structured testing, and controlled interaction strategies. Across the cases, the thesis identifies recurring patterns for reliable conversational systems, including deterministic data capture, controlled routing, structured fallbacks, document-grounded answering, and human handover support. Findings show that effectiveness in public administration depends less on model sophistication alone and more on system-level design, interactional robustness, governance mechanisms, and user experience quality. Conversational AI demonstrates potential to enable services that are more intuitive, responsive, and human-centred, reshaping the relationship between citizens and institutions. The thesis also highlights the value and competences of a humanistic-technological profile combining linguistic, interactional, and ethical expertise with technical knowledge. Overall, the study frames conversational AI as a socio-technical infrastructure for public-service innovation.