What method can be used to develop an AI tool that generates email replies at a Swedish authority?

Artificial intelligence is a highly relevant field that affects society, businesses, and individuals alike. The development of AI models is progressing rapidly, and for organizations, it is crucial to understand the resources, approaches, and competencies needed for successful implementation. AI tools can contribute to increased efficiency and improved quality in certain tasks. While the private sector is making significant investments in harnessing the potential of AI, the public sector is lagging behind, likely due to different requirements for information management and differing priorities. This thesis explores how a simple AI model can be developed for a department within a Swedish authority and investigates the potential impact such a model may have on daily operations. The result was a language model designed to generate email replies in a helpdesk environment. The model is based on the existing Llama 3.2 and has been tuned with RAG and internal documents. A visual demo version was also created, featuring a simple graphical interface built using Python and Tkinter. In addition to the technical implementation, a qualitative interview study was conducted with employees at the authority. The interviews were based on the Technology Acceptance Model (TAM) framework and showed that usability and user-friendliness are considered crucial for the use of the model. Information security and decreasing competence were highlighted as key factors. The study also indicates that skills development and training in AI and language models are essential to achieve acceptance and a successful implementation.