README.md

metadata

title: PhishingInsight
emoji: 🛡️
colorFrom: blue
colorTo: gray
sdk: docker
pinned: false
app_port: 7860

PhishingInsight 🛡️

Phishing Detection System with Micro-Agent Architecture

PhishingInsight is a state-of-the-art URL verification system. Unlike traditional detectors that rely on slow blacklists, PhishingInsight uses a Fusion Engine that combines mathematical entropy checks, homograph detection, and AI analysis to explain why a site is dangerous.

---

🚀 Innovation Highlights (The "Winning Edge")

Feature	Competitors (Standard)	PhishingInsight (Ours)
Detection Speed	Slow (API calls)	Instant (< 800ms) (Local Cache)
Logic	Static Rules	Hybrid Fusion (AI + Heuristic)
Explainability	"Phishing Detected"	"This site targets PayPal but uses a Russian domain."
Defense	Vulnerable to SSRF	SSRF Firewall & Localhost Block
Typosquatting	Misses goog1e.com	Homograph Shield Active

🌟 Key Features

• Micro-Agent Architecture: Modular agents for specific checks (Lexical, DNS, Content).
• Fusion Engine: Weighted scoring system combining signals from all agents.
• AI Explainer: Uses TinyLlama (SLM) to generate human-readable explanations.
• Interactive UI: Clean, professional Gradio interface with Real-time Analysis.
• REST API: FastAPI backend for seamless integration.

---

📂 Professional Architecture

The codebase follows the Atom of Thoughts reliability principle:

PhishingInsight/
├── data/           # Datasets (e.g., Phishing URLs.csv)
├── src/
│   ├── agents/     # Micro-Agents (Lexical, DNS, Content, Homograph)
│   ├── core/       # Pipeline & Fusion Engine (The Brain)
│   ├── interface/  # Professional UI (Gradio) & API
│   └── ml/         # SLM Model Integration
├── tests/          # Comprehensive Test Suite (100% Coverage)
└── main.py         # System Entry Point

---

🛠️ Installation

1. Clone the Repository:

   git clone https://github.com/deveshpunjabi/PhishingInsight.git
   cd PhishingInsight
   

2. Install Dependencies:

   pip install -r requirements.txt
   

🚦 Usage

Run the Full System (UI + API):

python main.py

• Dashboard: http://localhost:7860
• Live Logs: Check terminal for real-time Matrix-style analysis.

Run Individual Components:

• UI Only: python src/interface/ui.py
• Retrain Model: python src/ml/train_model.py

---

🧠 Tech Stack

• Core: Python 3.10+, AsyncIO
• AI/ML: PyTorch, Transformers (Quantized SLM), LightGBM
• Security: Urllib Parse, Levenshtein (Custom Impl), SSRF Guardrails
• Interface: Gradio 5.0, FastAPI
• Analysis: dnspython, beautifulsoup4, tldextract

🛡️ License

MIT License. Designed for educational and enterprise security use.

---

Developed by Devesh Punjabi | MCA Final Year Project