A delivery manager developed an automated workflow to generate meeting minutes in Microsoft Teams using Power Automate and an LLM API. The process reduces documentation time from 30–40 minutes per meeting to 30 seconds of manual effort plus generation time. The system saves approximately 1.5–2 hours of work daily for 3–5 meetings.
The workflow triggers when a VTT transcript file is saved to a OneDrive folder. Power Automate sends the content to an LLM, which structures the text into a summary, action items, and notes before posting the result to a Teams channel. The author chose VTT format because DOCX files contain binary data that causes errors during API transmission.
Attempts to fully automate the process via Microsoft Graph API were unsuccessful due to platform limitations. Specifically, the API does not support automatic transcript retrieval for standard calendar meetings, and filtering online meetings remains restricted. The author spent 6 hours attempting to bypass these constraints before opting for the manual VTT upload.
The project was implemented through collaborative work with the Claude AI assistant. The author notes that selecting the right model is critical, as some LLMs include unnecessary reasoning blocks. The current setup uses GPT-5.6 Luna, which handles multilingual transcripts and extracts action items into tables effectively.
