Automated Research Paper Podcast Generator

Daily Serverless Research Paper Discovery and Podcast Synthesis using Semantic Scholar API and Google NotebookLM.

An automated, serverless Python pipeline that discovers high-impact research papers using the Semantic Scholar Recommendations API, ingests them into Google NotebookLM (notebooklm-py), triggers long-form structured audio overview podcast synthesis, and runs headlessly on GitHub Actions.

---

Overview

This project automates daily research paper discovery and audio overview creation. It evaluates research paper candidates based on user-defined seed paper embeddings and credibility thresholds, filters out uncited or irrelevant works, and instructs Google NotebookLM AI hosts to generate structured technical podcast episodes.

---

Key Features

• Semantic Scholar Recommendation Engine: Uses seed paper embeddings (Bayesian Optimization, Hyperparameter Tuning, Cognitive Neuroscience, Sleep and Memory) to discover relevant, open-access research papers.
• Credibility and Citation Filtering: Enforces minimum citation thresholds and recent publication filters to exclude low-quality or zero-citation papers.
• Dynamic Non-Repeating Pipeline: Shuffles seed subsets and maintains persistent recommendation history to guarantee new, distinct paper recommendations on every run.
• Structured Podcast Prompting: Directs NotebookLM hosts to follow a strict format: Abstract explanation, Conclusion and implications, and key Methodology details.
• NotebookLM Source Discovery Fallback: If direct paper URL ingestion fails (or DOI links are likely to be rate-limited), NotebookLM web research is used to discover and import alternative sources automatically.
• Automated Local Cookie Sync: Includes a native macOS CLI utility (sync_auth.py) that extracts and decrypts local Chromium/Dia browser session cookies via macOS Keychain and updates GitHub Actions Secrets automatically.
• Serverless & Cost Efficient: Executes on GitHub Actions in fire-and-forget mode, triggering cloud audio rendering in seconds without wasting runner compute.

---

Repository Structure

├── config.json                            # Seed papers, topics, minimum citations, and podcast prompt
├── paper_recommender.py                   # Semantic Scholar paper discovery and filtering engine
├── podcast_generator.py                   # NotebookLM ingestion and audio generation runner
├── sync_auth.py                           # Native macOS browser cookie extraction & GitHub secret sync
├── requirements.txt                       # Python dependencies
└── .github/workflows/paper_podcast_cron.yml  # GitHub Actions cron workflow (Daily 8:00 AM IST)

---

System Architecture

[ Seed Papers / Topics ]
           │
           ▼
[ Semantic Scholar API ] ──(Filters: Open Access PDF + Min Citations)──► [ Target Paper ]
                                                                               │
                                                                               ▼
[ GitHub Actions / Cron ] ──(Restores NOTEBOOKLM_STORAGE_STATE)────────► [ Google NotebookLM ]
                                                                               │
                                                                               ▼
                                                                 [ Audio Overview Podcast ]

---

Quick Start Guide

1. Prerequisites

• Python 3.11 or higher
• GitHub CLI (gh) authenticated via gh auth login

2. Install Dependencies

pip install -r requirements.txt

3. Local Authentication Sync & Automated Token Refresh

Run the local authentication utility to extract your session cookies and sync them to your GitHub repository secret (NOTEBOOKLM_STORAGE_STATE):

python3 sync_auth.py

To ensure your GitHub Actions secret never expires, install the local 12-hour automated sync cron job on your Mac:

bash setup_local_cron.sh

4. Scheduled and Manual Execution

• Scheduled: Runs automatically every day at 8:00 AM IST (02:30 UTC).
• Manual: Trigger on-demand via the GitHub Actions UI under Actions > Generate Research Paper Podcast > Run workflow.

---

Configuration

Edit config.json to customize your seed papers, topic preferences, and citation filters:

{
  "seed_papers": [
    {
      "title": "Practical Bayesian Optimization of Machine Learning Algorithms",
      "id": "ARXIV:1206.2944"
    },
    {
      "title": "Sleep-dependent learning: a nap to remember",
      "id": "DOI:10.1038/nn1063"
    }
  ],
  "min_citations": 5,
  "podcast_format": "deep-dive"
}

---

License

This project is open-source software licensed under the MIT License.