RAG Security Sandbox 🛡️🤖

📋 Executive Summary

The RAG Security Sandbox is a professional DevSecOps playground and QA automation suite designed to validate the security posture of Retrieval-Augmented Generation (RAG) applications. It demonstrates practical AI QA/security engineering by simulating modern attacks (Prompt Injection, Data Poisoning, Prompt Leaking) and proving the efficacy of multi-layered AI defenses.

---

📸 Automated DevSecOps Workflow in Action

1. Automated Security Fuzzing

Executing adversarial payloads via Pytest against secure and vulnerable API endpoints.

2. Live SOC Monitoring

Real-time observability of AI decisions, latencies, and active threats via custom Tailwind CSS dashboard.

3. Executive AI Risk Reporting

Automated CISO-level HTML report generated post-fuzzing, summarizing vulnerabilities and DLP guardrail effectiveness.

---

🌟 Key Features

• True RAG Architecture: Uses the ChromaDB vector database for realistic, production-like context retrieval and knowledge grounding.
• Dual Security Guardrails: Implements a defense-in-depth strategy using Regex for lightning-fast deterministic filtering, alongside an advanced "LLM-as-a-Judge" for deep semantic analysis and Data Loss Prevention (DLP).
• Automated QA & Reporting: The pytest test suite automatically triggers adversarial fuzzing and concludes with an auto-generated HTML Executive Risk Report outlining the system's security posture.
• Live SOC Dashboard: Features a real-time UI built with Tailwind CSS to visually monitor attacks, security decisions (ALLOW/BLOCK), and latencies.

🚀 Quick Start

Follow these steps to initialize the environment and start the security simulation:

# Create a virtual environment
python -m venv venv

# Activate the environment (Windows)
.\venv\Scripts\activate
# For Linux/macOS: source venv/bin/activate

# Install dependencies
pip install -r requirements.txt

# Initialize the Vector Database (ChromaDB)
python init_db.py

# Start the FastAPI Server
python app.py

🎯 Running the Attack Simulation

Once the FastAPI server is running (python app.py), you can trigger the automated adversarial fuzzer and observe the results.

Open the SOC Dashboard:

Navigate your browser to http://localhost:8000/dashboard to view the live Security Operations Center interface.

Run the Fuzzer:

In a separate terminal (with the virtual environment activated), execute the security test suite:

pytest tests/

Review the Results:

Watch the dashboard update in real-time. Once the tests complete, open the newly generated executive_risk_report.html in your browser to view the CISO summary.

---

Disclaimer: This project is for educational and security research purposes only. Always use LLMs and RAG systems responsibly!