metadata

title: ZeroPhish Gate
emoji: 🐠
colorFrom: blue
colorTo: gray
sdk: gradio
sdk_version: 5.38.2
app_file: app.py
pinned: false
license: mit
short_description: Zero Trust. Zero Phishing. Zero Compromises

🛡️ ZeroPhish Gate - AI-Powered Phishing Detection

Zero Trust. Zero Phishing. Zero Compromises.

An advanced AI-powered security analysis tool that combines BERT and LLaMA models to detect phishing, spam, malware, and social engineering attacks in emails and messages.

🌟 Features

🤖 Hybrid AI Analysis

BERT Model: Technical pattern detection for known phishing indicators
LLaMA Model: Semantic analysis for context understanding and social engineering detection
RAG-Based Reranking: Advanced reasoning that combines both models for superior accuracy

🛡️ Security Capabilities

Phishing Detection: Advanced identification of credential theft attempts
Social Engineering: Recognition of psychological manipulation tactics
Malware Analysis: Detection of suspicious links and attachments
Business Email Compromise: Identification of CEO fraud and wire transfer scams
Multi-Format Support: Analyze emails, SMS, and document attachments (PDF/TXT)

🌍 Enterprise Features

8+ Languages: Full analysis and responses in multiple languages
Role-Based Guidance: Tailored security advice for different job functions
Threat Scoring: 0-100 risk assessment with color-coded indicators
Detailed Reports: Downloadable security analysis reports
Audio Summaries: Text-to-speech for accessibility
Analysis History: Track and review previous security checks

🚀 How It Works

Three-Stage Analysis Pipeline

Stage 1 - BERT Pattern Detection
- Analyzes message structure and content patterns
- Identifies technical phishing indicators
- Provides confidence scoring based on learned patterns
Stage 2 - LLaMA Semantic Reanalysis
- Understands context and intent behind messages
- Detects social engineering and manipulation tactics
- Performs advanced reasoning about message content
Stage 3 - Hybrid Decision Making
- Combines insights from both AI models
- Reduces false positives and negatives
- Generates human-readable explanations

🎯 Threat Classification

Score Range	Level	Description
0-20%	🟢 Safe	No threats detected
21-40%	🟡 Low Risk	Minor security concerns
41-60%	🟠 Medium Risk	Requires careful attention
61-80%	🔴 High Risk	Likely security threat
81-100%	⚫ Critical	Immediate action required

🔧 Setup Instructions

For Hugging Face Spaces Deployment

Create a new Space on Hugging Face with Gradio SDK
Add required secrets in Space settings:
- GROQ_API_KEY: Your Groq API key for LLaMA access
- HF_TOKEN: Your Hugging Face token for model access
Upload files:
- app.py (main application)
- requirements.txt (dependencies)
- README.md (this file)

Required API Keys

Groq API Key: Sign up at console.groq.com for LLaMA model access
Hugging Face Token: Get from huggingface.co/settings/tokens

🎨 Professional Interface

The application features a modern, ChatGPT-inspired interface with:

Gradient Headers: Professional visual design
Card-Based Layout: Clean, organized sections
Responsive Design: Works on desktop and mobile
Real-time Analysis: Live feedback during processing
Interactive Elements: Hover effects and smooth animations
Accessibility: Screen reader support and keyboard navigation

🔒 Security & Privacy

Zero Data Storage: Messages are not permanently stored
Encrypted Processing: All analysis in secure environment
No User Tracking: Privacy-focused design
Temporary Files: Reports auto-deleted after download

🌐 Multi-Language Support

Full analysis and user interface support for 45+ languages including:

English, Spanish, French, German, Italian
Arabic, Urdu, Hindi, Bengali, Punjabi
Chinese, Japanese, Korean, Thai, Vietnamese
Russian, Ukrainian, Polish, Czech, Slovak
Portuguese, Dutch, Swedish, Norwegian, Danish
And many more...

🏢 Industry Applications

Supply Chain Security

Procurement: Verify vendor communications
Logistics: Analyze shipping notifications
Finance: Detect invoice and payment scams
Warehousing: Check delivery confirmations
Administration: General email security

Use Cases

Employee security training
Real-time threat assessment
Security awareness programs
Incident response support
Compliance documentation

📊 Technical Specifications

Models: BERT (ealvaradob/bert-finetuned-phishing) + LLaMA 3-8B
Languages: 45+ supported languages
File Formats: PDF, TXT, direct text input
Response Time: < 30 seconds average
Accuracy: 95%+ with hybrid analysis
Deployment: Hugging Face Spaces, Gradio

🤝 Contributing

This is a professional security tool. For feature requests or bug reports, please contact the development team.

📄 License

🆘 Support

For technical support or security questions:

Check the in-app help section
Review the glossary for term definitions
Contact your IT security team for organization-specific guidance

⚠️ Security Notice: This tool is designed to assist with security analysis but should not replace human judgment in critical security decisions. Always verify suspicious content through multiple channels and follow your organization's security protocols.