This thesis bridges the gap between complex AI regulations and practical implementation needs for AI practitioners navigating the EU AI Act. Initial research revealed that while extensive regulatory resources exist, they remain fragmented and difficult for non-legal practitioners to understand and apply effectively. As a result, the research focused on developing a conversational AI tool using multi-agent systems and LLM fine-tuning to make regulatory guidance more accessible and actionable.

To validate these findings, expert interviews were conducted with legal professionals, user surveys gathered practitioner needs, and comprehensive user testing evaluated the AI Risk Navigator prototype. Additionally, benchmarking was performed comparing the AI tool's performance using customized agent evaluation framework in providing accurate compliance guidance. The findings showed that the conversational AI significantly improves regulatory understanding among AI practitioners by translating complex legal language into more accessible guidance, though the practical development guidance remained somewhat generic, while decision-making confidence showed neutral results.

However, the study identified limitations in handling highly specific use cases and providing context-tailored advice, emphasizing the need for AI-generated guidance to be complemented with professional legal consultation for critical compliance decisions. The relationship between AI tools and human expertise should thus be complementary to achieve optimal regulatory navigation results. ... Read more