Bookify

Convert a YouTube playlist into a structured, citation-backed technical PDF book.

What it produces

• Topic-grouped chapters (not one chapter per video)
• LLM-written prose with citation verification
• Introduction, conclusion, glossary, references
• Final output: output/book.pdf

Current pipeline architecture

flowchart TD
    A[Playlist URL] --> B[Stage 1: Fetch metadata/audio + reference URLs]
    B --> C[Stage 2: Transcripts from YouTube Transcript API]
    C --> D[Stage 3: Terminology correction]
    D --> E[Stage 4: Group + order topics]
    E --> F[Stage 5a: Write topic chapters]
    F --> G[Stage 5b: Verify citations]
    G --> H[Stage 5c: Polish prose]
    H --> I[Stage 6: Assemble full book markdown]
    I --> J[Stage 7: Render PDF]
    J --> K[output/book.pdf]

Loading

Setup

1. Create venv and install:

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

2. Add API key in .env:

GEMINI_API_KEY=your_key_here

3. Configure provider/model in config.yaml (default is Gemini Flash).

Run

Full run:

DYLD_LIBRARY_PATH=/opt/homebrew/lib python run.py --playlist "https://www.youtube.com/playlist?list=YOUR_LIST_ID"

Resume from stage:

DYLD_LIBRARY_PATH=/opt/homebrew/lib python run.py --playlist "..." --from 3

Re-render PDF only:

DYLD_LIBRARY_PATH=/opt/homebrew/lib python run.py --from 7 --to 7

Key config (current)

llm:
  provider: gemini
  model: gemini-flash-latest
  temperature: 0.3

pipeline:
  batch_size: 4
  rate_limit_rpm: 6
  min_words_per_topic: 8000

Checkpoints

Important directories:

• checkpoints/01_fetch
• checkpoints/01b_ref_content
• checkpoints/02_transcripts
• checkpoints/02b_corrected
• checkpoints/03_groups
• checkpoints/04_topics
• checkpoints/04b_verified
• checkpoints/04c_polished
• checkpoints/05_book

Audio cache is local-only and ignored from git:

• checkpoints/audio/

Submission notes

• .claude/ is ignored and should not be committed.
• checkpoints/audio/ is removed/ignored.
• Checkpoints and generated book artifacts are versioned as required for reproducible runs.

Project structure

Bookify/
├── run.py
├── config.yaml
├── requirements.txt
├── README.md
├── pipeline/
│   ├── fetcher.py
│   ├── transcriber.py
│   ├── terminology_corrector.py
│   ├── grouper.py
│   ├── topic_writer.py
│   ├── citation_verifier.py
│   ├── prose_polisher.py
│   ├── assembler.py
│   └── pdf_renderer.py
├── llm/
│   └── client.py
├── utils/
│   ├── checkpoint.py
│   ├── progress.py
│   ├── quality_report.py
│   └── url_filter.py
├── checkpoints/
└── output/

Author

Aditya Chaurasiya
Indian Institute Of Technology Bombay