Large Language and Visual Assistant for User-Centric Cybersecurity

A sophisticated AI-powered system for detecting and analyzing potential scam content in images and text using advanced language models and computer vision techniques.

Features

• Real-time image and text analysis for scam detection
• Integration with Gemma 3:4B language model for advanced content understanding
• OCR capabilities for text extraction from images
• Configurable evaluation pipeline with customizable prompts
• Comprehensive logging and evaluation metrics
• Support for both image and text-only content analysis

FULL REPORT AVAILABLE IN REPORT.PDF

Installation

1. Clone the repository:

git clone https://github.com/yourusername/CSVA.git
cd CSVA

2. Create and activate a virtual environment:

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies:

pip install -r requirements.txt

4. Install Ollama (required for Gemma model):

# Follow instructions at https://ollama.ai/download

5. Pull the Gemma model:

ollama pull gemma:3b

Configuration

1. Review and modify config.yaml to adjust:
   • Model parameters
   • Evaluation prompts
   • Logging settings
   • Processing thresholds

Usage

1. Run the main evaluation pipeline:

python main.py

2. For testing and evaluation:

python test.py

Project Structure

• core/: Core evaluation pipeline components
• processors/: Image and text processing modules
• llms/: Language model integration
• utils/: Utility functions and helpers
• data/: Data storage and processing
• logs/: Evaluation and processing logs
• captures/: Screenshot storage
• captions/: Image caption storage