Research & Awareness Design
Year :
2025
Industry :
Digital Safety / AI Ethics
Client :
Self-Initiated Research
Project Duration :
3 months
Why This Project Matters
Today’s job market is being reshaped by generative AI — not just for good, but also for exploitation. Scammers now use advanced AI models to craft job postings that look polished, credible, and dangerously convincing. Job seekers lost $750.6M in 2024, and the traditional signs of “scam language” are no longer reliable.
This project explores a critical question:
Can machine-learning models still detect scams when the scammer is using AI too?
Our goal wasn’t only to analyze the data — it was to understand how AI changes digital trust and to translate these findings into simple, accessible resources for students and young professionals.
How I Investigated AI-Written Scams
Working with 17,880 real and fake job postings, we re-generated every fake post using an LLM to create modern, convincing scam examples.
This allowed us to test models against realistic, AI-enhanced deception instead of outdated scam language.
What I built and tested:
AI-expanded & refined dataset
Clean preprocessing & reproducible pipeline
NLP analysis (frequency, POS tagging, readability, topics)
ML classifiers including XGBoost
Visual comparisons of real vs. fake vs. AI-generated job posts
And to make the research accessible:
A narrated explainer podcast
A comic book for quick learning
A website organizing all findings
This combined technical accuracy with human-centered communication.
Link to GitHub Repository
More Projects
Research & Awareness Design
Year :
2025
Industry :
Digital Safety / AI Ethics
Client :
Self-Initiated Research
Project Duration :
3 months
Why This Project Matters
Today’s job market is being reshaped by generative AI — not just for good, but also for exploitation. Scammers now use advanced AI models to craft job postings that look polished, credible, and dangerously convincing. Job seekers lost $750.6M in 2024, and the traditional signs of “scam language” are no longer reliable.
This project explores a critical question:
Can machine-learning models still detect scams when the scammer is using AI too?
Our goal wasn’t only to analyze the data — it was to understand how AI changes digital trust and to translate these findings into simple, accessible resources for students and young professionals.
How I Investigated AI-Written Scams
Working with 17,880 real and fake job postings, we re-generated every fake post using an LLM to create modern, convincing scam examples.
This allowed us to test models against realistic, AI-enhanced deception instead of outdated scam language.
What I built and tested:
AI-expanded & refined dataset
Clean preprocessing & reproducible pipeline
NLP analysis (frequency, POS tagging, readability, topics)
ML classifiers including XGBoost
Visual comparisons of real vs. fake vs. AI-generated job posts
And to make the research accessible:
A narrated explainer podcast
A comic book for quick learning
A website organizing all findings
This combined technical accuracy with human-centered communication.
Link to GitHub Repository
More Projects
Research & Awareness Design
Year :
2025
Industry :
Digital Safety / AI Ethics
Client :
Self-Initiated Research
Project Duration :
3 months
Why This Project Matters
Today’s job market is being reshaped by generative AI — not just for good, but also for exploitation. Scammers now use advanced AI models to craft job postings that look polished, credible, and dangerously convincing. Job seekers lost $750.6M in 2024, and the traditional signs of “scam language” are no longer reliable.
This project explores a critical question:
Can machine-learning models still detect scams when the scammer is using AI too?
Our goal wasn’t only to analyze the data — it was to understand how AI changes digital trust and to translate these findings into simple, accessible resources for students and young professionals.
How I Investigated AI-Written Scams
Working with 17,880 real and fake job postings, we re-generated every fake post using an LLM to create modern, convincing scam examples.
This allowed us to test models against realistic, AI-enhanced deception instead of outdated scam language.
What I built and tested:
AI-expanded & refined dataset
Clean preprocessing & reproducible pipeline
NLP analysis (frequency, POS tagging, readability, topics)
ML classifiers including XGBoost
Visual comparisons of real vs. fake vs. AI-generated job posts
And to make the research accessible:
A narrated explainer podcast
A comic book for quick learning
A website organizing all findings
This combined technical accuracy with human-centered communication.
Link to GitHub Repository