AISG National AI Challenge 2024 - Top 3 Winner

Project Overview

Developed an intelligent email fraud detection system that integrates directly with Gmail to identify phishing attempts and fraudulent emails.

Architecture

Backend (LangChain)

• Engineered Chain-of-Thought prompts for GPT-3.5-turbo
• Implemented sophisticated fraud detection logic
• Designed custom prompt templates for high accuracy

Frontend (Chrome Extension)

• Integrated Gmail API for seamless email access
• Real-time analysis of incoming emails
• User-friendly interface for fraud alerts

Key Features

Fraud Detection Elements

1. Unrealistic demands and threats
2. Poor spelling and grammar
3. Suspicious URLs and links
4. Requests for confidential information
5. Vague or generic salutations
6. Improper domain names
7. Inconsistent sender details

Technical Implementation

Used JSON Output Parser from LangChain to structure responses:

• is_phishing (boolean)
• phishing_score (0-10 scale)
• rationale (detailed explanation)

Impact

The system achieved high accuracy in detecting various types of email fraud while minimizing false positives, making email communication safer for users.

GitHub Repository