This master's thesis proposes to enhance AI-generated meeting minutes by leveraging structured prompting and schema-based strategies within a transcript-only setup. Current AI models often produce summaries that lack actionable details such as tasks, decisions, and assigned responsibilities, which limits their effectiveness in organizational settings. The research aims to address this gap by exploring how prompt engineering and design constraints can improve the clarity and usefulness of automatically generated meeting minutes.

Utilizing a publicly available dataset of agile meeting ceremonies, the study will produce three versions of meeting minutes: human-written, AI-generated using a generic prompting baseline (ChatGPT), and AI-generated using a custom-developed prototype with structured prompting and schema alignment. A mixed-methods approach will be adopted, involving qualitative interviews with individuals experienced in meeting documentation to identify existing challenges, and qualitative evaluations by participants familiar with formal meetings to assess the effectiveness of the AI-generated minutes.

By comparing these different versions, the research seeks to determine whether prompting strategies and schema-guided output formats enhances the AI's ability to produce clear, complete, and actionable meeting summaries. The findings are expected to contribute to both academic knowledge and practical applications, offering insights into improving AI-assisted meeting documentation. This could lead to increased efficiency and accountability in organizations by leveraging AI technologies to generate high-quality meeting minutes from transcript data.