Honey-Pot
An AI-powered security tool that engages scammers with an autonomous agent to extract actionable intelligence. Using Google Gemini, it automatically identifies bank details and phishing links in real-time conversations.
Executive Summary
Passive spam filters fail to stop social engineering. This system actively engages threats to extract intelligence.
The Agentic Honey-Pot System represents a proactive shift in cybersecurity. It uses LLMs to gather actionable intelligence that can be used to track and shut down scam operations.
Core Infrastructure
AI-Powered Detection
Real-time analysis using Google Gemini with confidence scoring.
Autonomous Engagement
AI agent that mimics a naive user to keep scammers engaged.
Automatic Intel Extraction
Identifies bank details and phone numbers from conversations.
Live Threat Dashboard
Real-time visualization of detected threats.
Design Philosophy
I wanted to turn the tables on scammers. Instead of just blocking them, the vision was to use an AI agent to waste their time while extracting UPI IDs and bank details.
The realization that Google's Gemini could act as a 'digital decoy,' automatically extracting structured threat data from unstructured, malicious conversations.
Technical Architecture
Deduplicating extracted intelligence in real-time and maintaining a convincing 'naive user' persona for the AI agent during adversarial interactions.
Engineered With
- React 19
- TypeScript
- Google Gemini AI
- Tailwind CSS
- Vite
- Lucide React
Performance Goal
- Real-time LLM response streaming
- Low-latency threat detection
- Efficient concurrent session management
System Integrity
- Secure API key management
- Anonymized threat intelligence extraction
- Adversarial-resistant agent logic
