Gemini API + Python automate meeting minutes to JSON

The post frames the problem plainly: meeting audio holds important information, but transcribing, summarizing and drafting minutes by hand consumes significant time and invites missed details or misreadings. The author, who already automates work with AI and RPA, treats that manual effort as the thing to remove.

The pipeline is split into two stages connected by a text file. First, audio is uploaded to Google Cloud Storage and handed to gemini-pro for transcription, with the script polling while the file is processed and saving the transcript to transcript.txt. Second, a second script passes that transcript back to Gemini with a prompt that specifies a JSON shape — summary, decisions, and ToDo entries with assignee and due date — and writes the parsed result to meeting_report.json.

The author is candid about where it can break. Transcription quality shifts with room acoustics, voice and background noise; long recordings may need splitting or a speech-specific service; and output quality depends heavily on prompt wording, which is why the example spells out the JSON format. API rate limits and usage costs are flagged as operational concerns for high-volume use. Beyond minutes, the author suggests the same approach could support sales-call analysis and knowledge sharing.