🛡️ SecuRAG: Secure RAG Agent with Hybrid Defense

A Master's Project Prototype for Multimedia Security & Adversarial Defense

(Please replace with your actual diagram file name)

📖 Introduction

SecuRAG is a Retrieval-Augmented Generation (RAG) system designed with a "Security-First" approach. Unlike traditional RAG systems, SecuRAG implements a Hybrid Defense Layer to filter adversarial prompts (e.g., Jailbreak attacks) before they reach the LLM or Knowledge Base.

Key Features

• Hybrid Security Layer: Combines deterministic Regex filters with a semantic AI Firewall (DeepSeek/Llama3).
• Local Privacy Mode: Fully functional offline using Ollama and local ChromaDB. No data egress.
• Dynamic Knowledge Base: Supports PDF ingestion via a decoupled API.
• User-Friendly UI: A Streamlit-based dashboard with real-time audit visualization.

---

Architecture

The system follows a Microservices Architecture:

1. Frontend: Streamlit (Port 8501) - User Interface.
2. Backend: FastAPI (Port 8000) - API Gateway & Business Logic.
3. Engine: SecuRAG Engine - Orchestrates Ollama and ChromaDB.
4. Database: ChromaDB (Persistent SQLite) - Vector storage.

---

Installation & Setup

Prerequisites

• Python 3.10+
• Ollama installed and running.
• Model pulled: ollama pull deepseek-r1:1.5b (or your chosen model).

1. Clone & Install Dependencies

# Install required Python packages
pip install -r requirements.txt