Thesis title: LLM-Based Tools for Workers in the Public Sector
The aim of this doctoral project is to explore how LLMs can redefine machine support for knowledge workers in the public sector.
The research developed through four sequential phases. In the first, adoption readiness, the study analyzed the willingness of public knowledge workers to collaborate with LLM-based tools, highlighting how curiosity, critical thinking, and organizational context influence their propensity to adopt such technologies. The second phase, technical feasibility, assessed the potential use of LLMs in complex tasks such as judicial decision drafting, revealing the advantages of specialized models but also their high costs and limited sustainability. In the third phase, practical feasibility, a hybrid approach was developed that combines LLMs and symbolic representations to ensure transparency and control in public processes. Finally, in the last phase, final paradigm, building on the experience gained in the previous stages, a general paradigm was defined for introducing LLM-based tools in the public sector, grounded on four key principles: early expert involvement, readiness for use, modular design, and ease of intervention.