Quantum RAG Benchmark - Agriculture

Comprehensive benchmark comparing Quantum-Enhanced RAG vs Classical RAG for agricultural information retrieval using real-world data from 50+ trusted sources.

Project Structure

Quantum-Rag-Benchmark-agri/
├── web_crawler.py              # Scrapes 50+ agricultural sources
├── quantum_rag.py              # Quantum-enhanced RAG system
├── classical_rag.py            # Classical RAG baseline
├── compare_rag_results.py      # Automated comparison script
├── run_pipeline.py             # Master pipeline runner
├── requirements.txt            # Python dependencies
├── setup_crawler.sh            # Crawler setup script
│
├── agricultural_data_complete/ # Web crawler output
│   ├── txt/                    # Clean TXT files (50+ sources)
│   ├── json/                   # Metadata files
│   └── logs/                   # Scraping statistics
│
├── old/                        # Original implementation
│   └── src/
│       ├── quantum_embeddings/ # Quantum feature maps
│       ├── quantum_rag.py      # Original quantum RAG
│       └── baseline_rag.py     # Original baseline
│
└── *.csv, *.json               # Results and logs

Features

🌐 Web Crawler

• 50+ Trusted Sources: FAO, USDA, World Bank, CGIAR, research institutions
• Clean TXT Output: Perfect for RAG systems
• Organized Storage: Separate folders for txt, json, logs
• Automatic Merging: Creates unified corpus file

⚛️ Quantum RAG

• Quantum Feature Maps: Angle, Amplitude, IQP embeddings
• PennyLane & Qiskit: Multiple quantum backends
• Hybrid Embeddings: Combines classical + quantum features
• Configurable Qubits: 4-16 qubits supported

📊 Classical RAG (Baseline)

• MiniLM Embeddings: Fast, 384-dimensional
• Qdrant Vector DB: In-memory for speed
• T5 Generation: Local answer generation
• Gemini API Support: Optional cloud LLM

🔬 Comparison Framework

• Automated Benchmarking: 10 test queries
• Multiple Metrics: Speed, similarity, diversity, overlap
• Statistical Analysis: Aggregate results
• Export Formats: JSON and CSV

Quick Start

1. Install Dependencies

pip install -r requirements.txt
playwright install  # For web crawler

2. Run Complete Pipeline

python run_pipeline.py

This will:

1. Check dependencies
2. Run web crawler (if no data exists)
3. Run automated comparison
4. Generate results

3. Or Run Components Individually

# Step 1: Collect data (3-5 minutes)
python web_crawler.py

# Step 2: Run comparison
python compare_rag_results.py

# Step 3: Try interactive systems
python quantum_rag.py      # Quantum-enhanced
python classical_rag.py    # Classical baseline

Usage Examples

Web Crawler

# Full scraping (50+ sources)
python web_crawler.py

# Quick sample (10 sources for testing)
# Edit web_crawler.py and uncomment:
# asyncio.run(scrape_quick_sample())

Quantum RAG

python quantum_rag.py

Options:

• Embedding Type: angle, amplitude, classical
• Qubits: 4-16 (default: 8)
• Interactive Mode: Ask questions in real-time

Classical RAG

python classical_rag.py

Features:

• Fast classical embeddings
• Same interface as Quantum RAG
• Logs to CSV for comparison

Comparison

python compare_rag_results.py

Outputs:

• rag_comparison_results.json - Detailed results
• rag_comparison_results.csv - Spreadsheet format

Results & Metrics

The comparison evaluates:

1. Retrieval Speed

   • Average time per query
   • Speed ratio (quantum vs classical)

2. Retrieval Quality

   • Average similarity scores
   • Top-k overlap between systems

3. Source Diversity

   • Variety of sources in results
   • Coverage across corpus

4. Statistical Significance

   • Aggregate metrics
   • Per-query analysis

Configuration

Quantum Settings

In quantum_rag.py:

# Embedding types
- angle: Simple rotation-based (fast)
- amplitude: Dense state preparation (expressive)
- iqp: Instantaneous Quantum Polynomial (complex)

# Qubits
n_qubits = 8  # 4-16 recommended

Classical Settings

In classical_rag.py:

# Embedding model
embedder = SentenceTransformer("all-MiniLM-L6-v2")

# Chunk settings
chunk_size = 500  # words
overlap = 50      # words

Crawler Settings

In web_crawler.py:

# Delay between requests (be polite!)
delay = 3  # seconds

# Retry attempts
max_retries = 3

Data Sources

The web crawler collects from:

International Organizations (10)

• FAO (Food and Agriculture Organization)
• World Bank Agriculture
• CGIAR Research

US Government (8)

• USDA Farming, Crops, Livestock
• Economic Research Service
• National Agricultural Statistics

Other Governments (5)

• UK DEFRA
• EU Agriculture
• Australia, India

Research Institutions (10+)

• CIMMYT, IRRI, ICRISAT
• IFPRI, CSIRO

Other (7)

• AGRIS Database
• Precision Agriculture
• Sustainable Agriculture

Requirements

Python Packages

• qdrant-client >= 1.7.0
• sentence-transformers >= 2.2.2
• transformers >= 4.35.0
• crawl4ai >= 0.2.0
• pennylane >= 0.33.0
• qiskit >= 0.45.0
• numpy, pandas, tqdm

Optional

• google-generativeai (for Gemini API)
• python-dotenv (for .env support)

Hardware

• CPU: Any modern processor
• RAM: 8GB+ recommended
• Storage: 1GB for data
• GPU: Not required (CPU mode)

Gemini API (Optional)

For better answer generation:

1. Get API key from https://makersuite.google.com/app/apikey

2. Create .env file:

GEMINI_API_KEY=your-api-key-here

3. Run any RAG system - it will auto-detect Gemini

Troubleshooting

Import Errors

pip install -r requirements.txt
playwright install

No Data Found

python web_crawler.py
# Wait 3-5 minutes for completion

Quantum Import Fails

pip install pennylane qiskit
# Or use classical mode

Crawler Fails

• Check internet connection
• Some sites may block automated access
• URLs may have changed
• Check logs in agricultural_data_complete/logs/

Performance Tips

1. Use Quick Sample First: Test with 10 sources before full scraping
2. Classical for Speed: Use classical RAG for fast prototyping
3. Lower Qubits: Start with 4-6 qubits for faster quantum processing
4. Cache Models: Models are cached after first download

Project Timeline

1. Data Collection: 3-5 minutes (web crawler)
2. Indexing: 2-3 minutes (first time)
3. Comparison: 1-2 minutes (10 queries)
4. Interactive Use: Real-time queries

Total: ~10 minutes for complete pipeline

Results Format

JSON Output

{
  "timestamp": "2025-11-03T...",
  "num_queries": 10,
  "queries": [...],
  "aggregates": {
    "classical": {...},
    "quantum": {...},
    "comparison": {...}
  }
}

CSV Output

Columns: query, classical_time, classical_similarity, quantum_time, quantum_similarity, overlap, speedup, score_improvement_pct

Contributing

Contributions welcome! Areas:

• Additional data sources
• New quantum feature maps
• Evaluation metrics
• Visualization tools

License

See LICENSE file in repository

Citation

If you use this benchmark in research:

@software{quantum_rag_agri_2025,
  title={Quantum RAG Benchmark - Agriculture},
  author={...},
  year={2025},
  url={https://github.com/abeer555/Quantum-Rag-Benchmark-agri}
}

Support

• Issues: GitHub Issues
• Docs: See WEB_CRAWLER_README.md for crawler details
• Contact: Repository owner

---

Built with: PennyLane, Qiskit, Qdrant, Sentence Transformers, Crawl4AI