ZeroPhish Gate - AI-Powered Phishing Detection System Zero trust. Zero phishing. Zero compromises.
ZeroPhish Gate is a multilingual, AI-powered security assistant designed to detect and explain phishing, spam, and fraudulent content in emails, chats, or files. It combines traditional pattern-based techniques with semantic LLM reasoning to ensure both accuracy and human-friendly feedback.
Presentation + Demo Video: 8 minute Video
https://drive.google.com/file/d/1KXUa65tDNqBkxaN1tHW-zmxl3RFsl7y2/view?usp=sharing
Key Features Hybrid AI Threat Detection BERT-based pattern detection for known phishing traits LLaMA (via Groq API) for intent recognition and semantic analysis Retrieval-Augmented Generation (RAG) for reranked, context-rich results Multilingual & Accessible Supports 40+ languages Text-to-speech output via gTTS for accessibility User-Focused Output Input via plain text or uploaded PDF/TXT files Role-based actionable advice (e.g., for procurement, admin, finance) Glossary with hover-over terms for easy learning Downloadable security reports and summaries Risk Scoring & Feedback Visual badges with 5-tier risk level (0–100%) Interactive threat analysis history Audio summary of the scan for non-readers Quick Start Guide Paste or Upload:
Paste suspicious content OR upload a .pdf or .txt file.
Choose your preferred language and organizational role. Run Analysis:
Click "Analyze with AI" to evaluate for threats. Review & Act:
View detailed analysis with glossary and tips. Download a report or listen to a voice summary. Report suspicious content to IT. How It Works (Simplified) You enter suspicious content or upload a document. A BERT model checks for known phishing patterns. LLaMA interprets tone and context using natural language. Threat score is calculated, and advice is generated. You get results with clear visuals, definitions, and audio if needed. Technical Architecture Architecture Diagram
Diagram shows pipeline from input → analysis → scoring → output
Architecture (Mermaid version) Risk Scoring System Score Range Level Color Description 0–20 Safe 🟢 No threat detected 21–40 Minimal Suspicion 🟡 Minor concerns, safe to review 41–60 Needs Attention 🟠 Potential risk, review content 61–80 Likely Threat 🔴 High probability of phishing/spam 81–100 Severe Threat ⚫ Dangerous, report and avoid Glossary (Sample Terms) Hover over underlined terms in the app to learn more.
Local Installation
git clone https://huggingface.co/spaces/your-username/ZeroPhish-Gate cd ZeroPhish-Gate
pip install -r requirements.txt
python app.py
streamlit run ui_app.py Requirements Python 3.8+
Required packages:
gradio fitz # PyMuPDF gtts transformers python-dotenv groq Required API Keys Make sure to set the following in your .env file or environment variables:
colorFrom: purple colorTo: red sdk: docker
tags: streamlit pinned: false
short_description: Zero trust. Zero phishing. Zero compromises. license: mit
GROQ_API_KEY HF_TOKEN KAGGLE_USERNAME KAGGLE_KEY License MIT License – see LICENSE for details.
Contributions We welcome community contributions, issue reports, and feedback! Feel free to open a pull request or start a discussion.