Turn new podcast recordings into Notion content entries weekly

Build me an agent workflow that turns each new podcast or interview recording into a fully fleshed-out content entry in my Notion content library.

Trigger: cron, every Monday at 9am in my local timezone.

On each run, the agent should:

1. Call Google Drive List Files to find any audio files added to my 'Podcast Episodes' folder in the last 7 days. Use the Drive query language to filter on the folder's parent ID, a createdTime greater than 7 days ago, mimeType starting with audio/, and trashed = false. If there are no new files, exit cleanly without creating anything.

2. For each new audio file, call Google Drive Download File Content to fetch the raw bytes.

3. Call Deepgram Transcribe Pre-recorded Audio with the raw bytes as the request body. Enable diarization, paragraphs, smart formatting, and utterances. Use the nova-3 model. Keep the Deepgram response, the speaker-labeled transcript, and the paragraph timestamps for the next step.

4. From the transcript, generate: an episode title, a two-paragraph episode summary, a list of timestamped chapter markers derived from Deepgram's paragraph and topic boundaries (format MM:SS Chapter title), three pull quotes lifted verbatim from the transcript, and five social-ready snippets: one for LinkedIn, two for X (under 280 chars each), and two for the email newsletter (3 to 5 sentences each).

5. Call Notion Create a Page in my Content database. Map the generated fields to these properties: Title (episode title), Summary (the two-paragraph summary), Status (set to 'Draft'), Chapter Markers (rich text, one chapter per line), Pull Quote 1, Pull Quote 2, Pull Quote 3, LinkedIn Post, X Post 1, X Post 2, Newsletter Blurb 1, Newsletter Blurb 2, Source File (link to the Drive file), Recorded Date (the file's createdTime).

6. Call Notion Append Block Children on the page that was just created to add the full diarized transcript at the bottom. Use heading_2 'Full Transcript' followed by paragraph blocks, one per speaker turn, prefixed with the speaker label (e.g. 'Speaker 1: ...'). Keep paragraph blocks under Notion's 2000-character limit and split longer turns across multiple blocks.

7. Move on to the next audio file. Process them serially so a Deepgram failure on one episode doesn't kill the whole batch.

At the end of the run, log how many episodes were processed and any files that failed transcription.

I will provide: the Google Drive folder ID for 'Podcast Episodes', the Notion database ID for my Content library, and the exact property names in that database when I configure the workflow.