Daily arXiv Podcast Pipeline

This project creates a daily paper-to-audio workflow:

1. Fetch latest iArxiv daily email from Gmail.
2. Extract top-N ranked arXiv IDs from visible ranked rows.
3. Download PDFs directly from https://arxiv.org/pdf/<id>.pdf (tracking-safe, no iArxiv link usage).
4. Automate NotebookLM to upload an instructions.txt source plus papers, inject a custom Audio Overview prompt in Studio, then trigger audio overview.
5. Provide cleanup commands for local files and notebook deletion.

Quick start

1. Create env and install:
   • python3 -m venv .venv
   • source .venv/bin/activate
   • pip install -e .[dev]
   • playwright install chromium
2. Copy .env.example to .env and adjust paths/settings.
3. Place Gmail OAuth client secret file at GMAIL_CREDENTIALS_FILE.
4. For NotebookLM login prefill, copy harvardkey_credentials.example.json to harvardkey_credentials.json and set your credentials there (2FA still manual).
5. Run:
   • daily_podcast run
   • Test with sample email: daily_podcast run --eml-file EXAMPLE_IARXIV.eml
   • Note: run auto-cleans local downloaded files after notebook creation; it does not delete the notebook.

Commands

• daily_podcast run [--date YYYY-MM-DD] [--eml-file path/to/file.eml]
• daily_podcast fetch-email [--date YYYY-MM-DD] [--eml-file path/to/file.eml]
• daily_podcast download-pdfs [--date YYYY-MM-DD | --latest]
• daily_podcast create-notebook [--date YYYY-MM-DD | --latest]
• daily_podcast cleanup-files [--date YYYY-MM-DD | --latest]
• daily_podcast cleanup-notebook [--date YYYY-MM-DD | --latest]

For local testing without Gmail credentials, use:

• daily_podcast fetch-email --eml-file EXAMPLE_IARXIV.eml

Handoff Guide

• Full setup handoff doc (LLM-friendly): LLM_SETUP_HANDOFF.md