VerifAI – Verify Before You Trust.

Inspiration

The inspiration behind VerifAI came from the growing number of fraudulent job and internship opportunities targeting students and fresh graduates. With online recruitment becoming increasingly common, distinguishing genuine opportunities from scams is becoming harder. We wanted to build something that could help job seekers make safer and more informed career decisions before trusting an opportunity.

What it does

VerifAI is an AI-powered recruitment scam detection platform that analyses job postings, internship descriptions, and recruitment documents to identify potentially fraudulent opportunities.

It uses machine learning and rule-based detection to:

• Analyse job descriptions and uploaded PDF documents.
• Identify suspicious recruitment patterns and misleading claims.
• Detect warning signs such as registration fees, payment demands, and unrealistic promises.
• Generate a risk score to indicate the likelihood of a fraudulent opportunity.
• Provide explainable results highlighting the factors behind its assessment.

Rather than simply labelling an opportunity as genuine or fraudulent, VerifAI helps users understand why it may be suspicious.

How we built it

We built VerifAI using Python and FastAPI for the backend, SQLite for data storage, and Scikit-learn for machine learning.

Our detection model was trained using the Employment Scam Aegean Dataset (EMSCAD). We preprocessed the dataset, removed duplicate records, and applied TF-IDF vectorisation to convert job descriptions into numerical features.

A calibrated Linear Support Vector Machine (SVM) was used to classify recruitment postings. We complemented the model with rule-based checks to identify specific warning signs that may not be adequately captured by the machine learning model alone.

The backend integrates these components into an analysis pipeline that generates risk assessments and supporting explanations.

Challenges we ran into

• Data imbalance: Fraudulent job postings represent a smaller portion of the dataset, making reliable scam detection challenging.
• Feature extraction: Identifying meaningful patterns in unstructured job descriptions required careful text preprocessing and feature engineering.
• False positives and false negatives: We had to balance detecting suspicious opportunities without unnecessarily flagging legitimate job postings.
• Explainability: Providing meaningful reasons behind predictions was just as important as generating accurate classifications.
• Real-world variability: Recruitment scams constantly evolve, making it difficult for a model trained on historical data to recognise every emerging scam pattern.

Accomplishments that we're proud of

• Successfully developed a machine-learning-based recruitment scam detection system.
• Integrated ML predictions with rule-based detection for a more comprehensive assessment.
• Implemented both text-based and PDF-based job analysis.
• Developed explainable risk assessments instead of relying on simple binary predictions.
• Achieved an F1-score of 79.53% and a PR-AUC of 87.87% on our held-out evaluation data.
• Built a functional backend with automated tests to validate its core functionality.

Most importantly, we transformed a real-world problem affecting students and job seekers into a practical AI-driven solution.

What we learned

Through this project, we gained practical experience in building and evaluating machine learning pipelines, working with imbalanced datasets, feature engineering, model calibration, and integrating ML models into backend applications.

We also learned that building a reliable AI system involves more than achieving good evaluation metrics. Understanding model limitations, reducing false predictions, and making results interpretable are equally important when developing solutions that influence real-world decisions.

What's next for VerifAI

We envision expanding VerifAI into a more comprehensive recruitment safety platform.

Our future plans include:

• Company Verification: Verify company identities and cross-check publicly available company information.
• Recruiter Verification: Identify suspicious recruiter profiles and impersonation attempts.
• URL and Email Analysis: Analyse recruitment links, email addresses, and domains for potential phishing indicators.
• Multilingual Detection: Support recruitment content in multiple languages.
• Continuous Learning: Improve detection capabilities using newer scam patterns and verified feedback.
• Browser Integration: Enable users to assess suspicious job postings directly while browsing recruitment websites.
• Community Reporting: Allow users to report suspicious opportunities and contribute to a growing intelligence database.

Our long-term goal is to make recruitment safer by helping people identify risks before they become victims.

VerifAI – Verify Before You Trust.

Built With

• ai
• artificial-intelligence
• cybersecurity
• data-science
• deep-learning
• explainable-ai
• fastapi
• fraud-detection
• fraud-prevention
• internship-verification
• job-scam-detection
• linear-svm
• machine-learning
• natural-language-processing
• python
• recruitment-safety
• risk-assessment
• scam-detection
• scikit-learn
• social
• student-safety
• tf-idf
• trust-&-safety
• xai