Built to protect
Malaysian job seekers
JobShield is a web-based fake job scam detection system, developed as a final year project at Universiti Teknologi MARA (UiTM). It combines machine learning, scam-pattern rules and AI to help people spot job fraud before it costs them.
Job scams are a growing threat to Malaysian graduates.
Online platforms, social media and chat groups are now the main way people find jobs. They are also where fraudulent postings spread, leading to lost money, stolen identities and real distress.
Fresh graduates and students are the easiest targets. They often have not seen enough real job ads to notice the warning signs, and scam ads are written to look convincing and urgent.
From a job ad to a risk score
Whatever form the ad comes in, JobShield turns it into text first, then runs the same analysis on it.
Submit
Paste text, upload an image, enter a job link or scan a QR code. The Telegram bot accepts the same four inputs.
Extract the text
Images are read with OCR, QR codes are decoded, and job pages are fetched and cleaned so only the ad text remains.
Analyse
The NLP model, scam rules, structure check and location validation each give a score, combined into one risk score.
Explain
You see the risk level, the exact red flags found and AI advice on what to do next. Results are saved in your history.
Four layers, one score
A scam that slips past one check is likely to be caught by another. Each layer contributes a fixed share of the final risk score.
NLP model
TF-IDF turns the ad text into features. A Logistic Regression classifier trained on 17,880 EMSCAD job ads estimates how likely the ad is fraudulent.
Rule engine
Seven English scam checks: unrealistic salary, urgent language, upfront payment, personal information requests, no experience needed, vague description and suspicious links.
Structure check
Looks at how the ad is built: how detailed it is, whether key information is present, and whether the contact details look genuine.
Location validation
The Google Geocoding API checks that the job location is a real place. A location that does not exist is a warning sign.
Plus AI advice and a chat assistant
After the score, Groq's Llama 3.3 70B writes advice for the specific red flags found in your ad. The chatbot on every page answers questions about job scams. The AI explains the result but does not change the score.
How well does it work?
The NLP model was tested on 3,576 job ads it had never seen, 173 of them fraudulent. The training data has 17,880 ads, of which 4.8% are fraudulent.
Five-fold cross-validation F1: 0.7096 ± 0.0107.
The model is tuned to catch scams, even if that means some real ads get flagged too. For that reason JobShield works as a warning tool. Treat a high score as a reason to check the employer yourself, not as a final verdict.
Good to know
- •Works on job ads written in English.
- •Sites that need a login, such as Facebook and LinkedIn, cannot be opened. Paste the text instead.
- •OCR accuracy drops on busy, multi-column posters.
- •A result is guidance, not proof. Always verify the employer.
Built with modern tools
Siti Alya Hanisah Binti Latif
Ready to try it?
Paste a job ad, upload a screenshot, enter a link or scan a QR code, and get your result in seconds.
Start Free Analysis