🏛️ PolicyGraph: AI-Powered Policy Analysis Platform

Automated comparative analysis of international policy documents using Retrieval-Augmented Generation (RAG)

🎯 Problem Statement

Policy analysts at organizations like the IMF and World Bank spend hours manually comparing policy positions across hundreds of documents. PolicyGraph automates this process, reducing analysis time from 2 hours to 2 minutes.

✨ Key Features

• 🤖 Intelligent Comparative Analysis: Compare IMF vs World Bank perspectives on any policy topic
• 📊 Timeline Visualization: Track how policy focus evolves over time
• 📄 Auto-Generated Policy Memos: Export professional 3-page briefs in PDF format
• 🔍 Source Tracking: Every claim linked to specific documents and page numbers
• ⚡ Real-time Analysis: Powered by GPT-4 and semantic search
• 📈 Admin Dashboard: Monitor database health and add new documents

🏗️ Architecture

User Query → Smart Retrieval → Multi-Source Analysis → Structured Output
                ↓                      ↓                      ↓
         Vector Database         GPT-4 Reasoning      PDF/MD/Chart Export
         (Chroma DB)            (Comparative Logic)    (ReportLab/Plotly)

🚀 Quick Start

Prerequisites

• Python 3.10+
• OpenAI API Key
• 2GB disk space for vector database

Installation

# 1. Clone the repository
git clone https://github.com/yourusername/PolicyGraph.git
cd PolicyGraph

# 2. Create virtual environment
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# 3. Install dependencies
pip install -r requirements.txt

# 4. Set up environment variables
cp .env.example .env
# Edit .env and add your OPENAI_API_KEY

# 5. Add policy documents
# Place PDF files in ./docs/ folder

# 6. Build vector database
python ingest.py

# 7. Launch the application
streamlit run app.py

The app will open in your browser at http://localhost:8501

📂 Project Structure

PolicyGraph/
├── app.py                  # Main Streamlit application
├── ingest.py               # Document ingestion script
├── memo_generator.py       # PDF report generation
├── timeline_analyzer.py    # Temporal analysis module
├── admin_panel.py          # Database management UI
├── requirements.txt        # Python dependencies
├── .env.example            # Environment template
├── docs/                   # PDF documents folder
└── chroma_db/              # Vector database (auto-generated)

💡 Use Cases

For Policy Researchers

• Cross-Organization Comparison: "Compare IMF and World Bank views on climate finance"
• Gap Analysis: Identify topics covered by one organization but not another
• Trend Tracking: See how policy emphasis shifts over years

For Decision Makers

• Executive Briefs: Generate 3-page policy memos in seconds
• Source Verification: Every statement backed by document citations
• Multi-Perspective Analysis: Understand different organizational mandates

For Data Scientists

• RAG Pipeline Example: Production-ready retrieval-augmented generation
• Comparative AI: Techniques for balanced multi-source retrieval
• Document Analytics: Temporal and keyword analysis patterns

🛠️ Tech Stack

Component	Technology	Purpose
LLM	OpenAI GPT-4-mini	Natural language understanding
Embeddings	text-embedding-3-small	Semantic search
Vector DB	ChromaDB	Document storage & retrieval
Framework	LangChain	RAG orchestration
Frontend	Streamlit	Interactive UI
Visualization	Plotly	Charts & timelines
PDF Generation	ReportLab	Policy memo export

📊 Performance Metrics

• Documents Analyzed: 2 organizations, 312 chunks
• Average Query Time: < 3 seconds
• Retrieval Accuracy: 95%+ citation precision
• Comparative Balance: 50/50 IMF-WB coverage for comparison queries

🎓 Advanced Features

Smart Retrieval Algorithm

# Detects comparison queries and ensures balanced coverage
if "compare" in query:
    imf_results = search("IMF" + topic)
    wb_results = search("World Bank" + topic)
    return balanced_merge(imf_results, wb_results)

Timeline Analysis

• Extracts publication years from document metadata
• Tracks topic mention frequency over time
• Identifies trending vs declining policy areas

PDF Memo Generation

• Executive summary with key findings
• Side-by-side organizational comparison table
• Policy implications and knowledge gaps
• Full source citations

🗺️ Roadmap

• Multi-language support (French, Spanish, Arabic)
• Integration with IMF/WB APIs for real-time data
• Sentiment analysis across documents
• Automated policy contradiction detection
• Export to PowerPoint for presentations
• Email digest subscriptions

🤝 Contributing

Contributions welcome! Areas for improvement:

1. Add More Organizations: UN, OECD, ADB documents
2. Improve Chunking: Better document segmentation
3. Advanced Analytics: Network analysis, causal inference
4. UI Enhancements: Dark mode, accessibility features

Please read CONTRIBUTING.md for guidelines.

📝 License

MIT License - see LICENSE for details.

📧 Contact

Wuhao Xia

• LinkedIn: (https://www.linkedin.com/in/carlxia0902/)
• Email: (carlboy2002@gmail.com)
• Portfolio: (https://carlboy2002.github.io/))

🙏 Acknowledgments

• IMF and World Bank for open access policy documents
• OpenAI for GPT-4 API
• Streamlit community for amazing framework

---

Built with ❤️ for better policymaking

If you find this project useful, please consider giving it a ⭐ on GitHub!