With generative AI enabling scammers to create highly polished job postings, traditional scam detection methods are becoming ineffective. This project investigates whether machine-learning models—and humans—can still detect scams when deception itself is AI-generated.

In 2024 alone, job seekers lost over $750 million to recruitment scams. Many of these scams no longer contain obvious red flags like poor grammar or formatting, making first-time job seekers especially vulnerable.

To simulate modern scam behavior, we:

• Took a dataset of 17,880 job postings

• Regenerated all fraudulent postings using a large language model

• Applied NLP techniques (readability, frequency, POS tagging)

• Trained ML classifiers, including XGBoost, to evaluate detection performance

Key Insights

• AI-generated scams closely mimic legitimate job structures

• Traditional ML models often rely on superficial stylistic cues rather than true indicators of deception

• Overlap between real and fake postings is significantly higher than in older datasets

Human-Centered Communication

Recognizing that awareness is as important as detection, we translated our findings into accessible formats:

• A narrated educational podcast

• A comic book designed for quick, engaging learning

• An interactive website showcasing findings and a reproducible research pipeline

This project highlights both the limitations of current ML-based detection systems and the importance of human-centered communication in digital safety—especially for students and early-career professionals.

My Role (Primary Contributor Across the Project)

I was involved end-to-end, contributing to research design, analysis, interpretation, and communication of findings.

Project Resources

• Codebase & Data Pipeline: [GitHub Repository Link]

• Website: [Download Comic & Podcast]