Agentic RAG API (Runtime-only)

🚀 Advanced Agentic RAG with Zero-Hardcoding Intelligence

Next-generation RAG system featuring intelligent tool-calling, dynamic caching, hybrid retrieval fusion, and pattern-learning capabilities that remove the need for hardcoded logic through context-driven decision-making.

• 🤖 Agentic Tool Intelligence: HTTP requests, web search, API interactions) — fully context-driven decisions • ⚡ Smart Document Caching: Persistent chunk storage system providing 15–20× speed improvements for repeated queries with auto-cleanup and cache management • 🧠 Pattern Learning Engine: Dynamic mathematical and logical pattern extraction from documents — no explicit programming required • 🔍 Hybrid Retrieval Fusion: Semantic (BGE embeddings) + keyword (BM25) search with intelligent weight balancing and multi-language OCR • 🌐 Production-Ready Architecture: FastAPI backend, dual LLM fallbacks (Groq + Gemini), secure auth, full logging, and real-time monitoring

---

📋 Quick Setup Guide for Evaluators

Prerequisites

• Miniconda or Anaconda
• Internet connection for API calls
• Optional: GPU for faster embeddings (CPU supported)

---

🚀 Step 1: Clone & Setup

# Clone the repository
git clone <your-repo-url>
cd rag_hackrx

# Create a conda environment with Python 3.10
conda create -n rag_hackrx python=3.10 -y

# Activate the environment
conda activate rag_hackrx

# Install dependencies
pip install -r requirements.txt

# Install Tesseract OCR (including Malayalam support)
sudo apt-get install tesseract-ocr tesseract-ocr-mal -y

# Install additional tools
pip install tools frontend

---

🔑 Step 2: Environment Configuration

Create a .env file in the project root:

touch .env

Add your keys:

# Required: LLM API Keys
GROQ_API_KEY=your_groq_api_key_here
GOOGLE_API_KEY=your_gemini_api_key_here
LLAMA_CLOUD_API_KEY=your_llama_cloud_api_key_here

# Optional: OCR Support for non-English PDFs
TESSERACT_LANGS=eng+mal

Where to get keys:

• Groq → console.groq.com
• Google Gemini → ai.google.dev
• Llama Cloud → cloud.llamaindex.ai

---

▶️ Step 3: Start the Server

# Start the API server
python webhook_api.py

# Or using uvicorn
uvicorn webhook_api:app --host 0.0.0.0 --port 8000 --reload

When ready, you should see:

INFO:     Started server process
INFO:     Waiting for application startup.
INFO:     Application startup complete.
INFO:     Uvicorn running on http://0.0.0.0:8000

---

🧪 Testing the System

Method 1: Using curl

curl -X POST "http://localhost:8000/hackrx/run" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer hackrx2024" \
  -d '{
    "documents": "https://example.com/sample.pdf",
    "questions": [
      "What is the main topic of this document?",
      "Summarize the key points"
    ]
  }'

Method 2: Using the Test Script

python test_webhook.py

---

📊 Example API Usage

Basic Document Query

POST /hackrx/run
Headers: {
  "Authorization": "Bearer hackrx2024",
  "Content-Type": "application/json"
}
Body: {
  "documents": "https://arxiv.org/pdf/2301.00234.pdf",
  "questions": [
    "What is the paper about?",
    "What are the main contributions?",
    "What datasets were used?"
  ]
}

Mathematical Pattern Learning

{
  "documents": "https://example.com/math-rules.pdf",
  "questions": [
    "What is 25 + 17?",
    "Calculate 100 + 55"
  ]
}

Agentic Tool Calling

{
  "documents": "https://example.com/api-guide.pdf",
  "questions": [
    "What is the flight number for route XYZ?",
    "Get the current status from the API endpoint"
  ]
}

---

🏗 System Architecture

Core Pipeline

1. Document Input → URL validation & cache check
2. Smart Processing → LlamaParse or cached chunks
3. Agentic Decision → Tool calling when required
4. Hybrid Retrieval → Semantic + keyword fusion
5. Context-Aware Generation → Intelligent responses
6. Pattern Learning → Dynamic rule discovery

---

📁 Project Structure

rag_hackrx/
├── webhook_api.py       # Main API server
├── test_webhook.py      # Test script
├── requirements.txt     # Dependencies
├── .env                 # Environment variables
├── document_cache/      # Cached chunks
├── vector_store/        # ChromaDB storage
├── transaction_logs/    # Request/response logs
└── logs/                # System logs

---

🔍 Troubleshooting

Missing API keys → Check .env formatting (no extra spaces) Import errors → pip install -r requirements.txt --force-reinstall Port in use → uvicorn webhook_api:app --port 8001 Slow processing → First request builds cache; later ones are 20× faster Auth failed → Use Authorization: Bearer hackrx2024

---

🏆 Why This Stands Out

• 🚀 Zero hardcoding — AI-driven decisions
• ⚡ 20× faster with intelligent caching
• 🧠 Self-learning pattern recognition
• 🤖 Autonomous tool calling
• 🔍 Hybrid retrieval fusion
• 🌐 Production-ready