📚 AI Book Publisher — Multi-Agent Workflow with LangGraph, Gemini & Streamlit

A production-ready, multi-agent AI system that automates the book publishing pipeline using:

• 🧠 Google Gemini (Vertex AI) for rewriting & review
• 🔁 LangGraph for agent orchestration
• 💾 ChromaDB for version tracking and memory
• 📊 Streamlit UI for real-time human feedback
• ☁️ Cloud Run for scalable, serverless deployment

Automatically rewrite, review, revise, and approve content with AI — just provide a URL.

---

📝 Blog Walkthrough

📖 Read the full implementation guide on Medium:
🔗 AI-Powered Book Publication Workflow with LangGraph and Google Gemini

---

🚀 Features

Component	Functionality
🧠 Gemini AI	Rewrite & review content using Google’s LLMs
🔁 LangGraph	Multi-agent stateful flow with pause/resume capability
💾 ChromaDB	Store and search past versions by metadata
🧍 Human Review	Streamlit UI lets users approve, revise, or reject
☁️ Cloud Run	Fully serverless, autoscaling deployment

🗺️ System Architecture

flowchart TD
  _start_ --> writer_agent
  writer_agent --> reviewer_agent
  reviewer_agent --> manager_agent
  manager_agent -->|revision_needed| writer_agent
  manager_agent -->|approved| _end_
  manager_agent -->|quality_check| quality_check
  manager_agent -->|human_review| human_review
  human_review -->|approved| quality_check
  human_review -->|revision_needed| writer_agent
  human_review -->|rejected| _end_
  quality_check --> _end_

Loading

---

📥 How It Works

1. User inputs a URL

2. System scrapes content using Playwright

3. Writer agent rewrites the content (via Gemini)

4. Reviewer agent scores and comments

5. Manager agent routes it for:

   • ✅ approval
   • 🔁 revision
   • 🧍 human review
   • ✅ final quality check

6. Versioned outputs are saved in ChromaDB

7. System deployed via Docker + Cloud Run

---

🧑‍💻 Local Development

1. Clone the Repo

git clone https://github.com/your-username/ai-book-publisher.git
cd ai-book-publisher

2. Install Dependencies

pip install -r requirements.txt
playwright install chromium

3. Run the App

streamlit run main.py

---

☁️ Cloud Run Deployment

Build and Push Docker Image

docker build -t gcr.io/YOUR_PROJECT_ID/ai-book-publisher .
docker push gcr.io/YOUR_PROJECT_ID/ai-book-publisher

Deploy to Cloud Run

gcloud run deploy ai-book-publisher \
  --image gcr.io/YOUR_PROJECT_ID/ai-book-publisher \
  --platform managed \
  --region us-central1 \
  --allow-unauthenticated \
  --port 8080 \
  --cpu 2 \
  --memory 4Gi \
  --timeout 3600 \
  --max-instances 5

---

📂 Project Structure

.
├── main.py                  # Streamlit UI
├── book_workflow.py         # LangGraph stateful agent flow
├── agents/
│   ├── writer_agent.py
│   ├── reviewer_agent.py
│   └── manager_agent.py
├── chroma_manager.py       # Version storage
├── scraper.py              # Web content scraper (Playwright)
├── config.py               # Vertex AI + system settings
├── Dockerfile              # Cloud Run deployment
└── requirements.txt

---

🧪 Sample Workflow State

{
  "original_content": "...",
  "writer_output": "...",
  "reviewer_feedback": "...",
  "manager_decision": "revision_needed",
  "status": "awaiting_human_feedback",
  "iteration_count": 2
}

---

📦 Roadmap

• Vertex AI + LangGraph integration
• Human-in-the-loop feedback
• ChromaDB versioning
• Streamlit interface
• Cloud Run deployment
• Notion / Google Docs export
• Summarization + Chapter Planning agents

---

📺 Live Demo (Streamlit)

Experience the full multi-agent AI publishing pipeline in action through a clean, interactive Streamlit UI:

🔗 Try the Live Demo (replace with your URL)

✨ What You Can Do:

• 📝 Enter any public URL to scrape article content
• ✍️ Watch it get rewritten by the Writer Agent
• 🧠 Review AI feedback and decisions from other agents
• 🧍 Interact with human-in-the-loop approval buttons
• 🧾 View reviewer feedback and final quality checks

The demo also logs agent decisions and stores versioned content in ChromaDB in real time.

📄 License

MIT License © 2025 Sayan Hazra

---