Spot The Scam AI Job Fraud

Overview

In today's job market, the sophistication of scams has drastically increased, putting job seekers at risk. "Spot the Scam" emerges as a solution, offering a robust AI-driven fraud detection system specifically designed for job postings. By leveraging a combination of machine learning techniques and human oversight, this tool aims to enhance the safety and reliability of job boards and marketplaces.

What sets "Spot the Scam" apart is its commitment to precision and transparency. With high-precision detection mechanisms and a calibrated confidence scoring system, it enables users—from recruiters to educators—to act decisively and confidently against fraudulent postings. The interactive dashboard serves as an essential component, providing real-time scoring, reviews, and an AI-assisted analysis, making it an invaluable asset for safeguarding job seekers.

Features

• High-Precision Ensemble: Combines TF-IDF text features with tabular signals to ensure accurate detection of potential scams.
• Calibration-First Design: Utilizes Platt or isotonic calibration methods for dependable confidence scores, helping users trust the system's assessments.
• Flexible Model Options: Offers classical models like Logistic Regression and XGBoost, alongside the option for fine-tuning transformer models such as DistilBERT, catering to varying resource availability.
• Gray-Zone Policy: Ensures scores that fall within uncertain ranges are handed off to manual review, maintaining a balanced approach to false positives.
• Transparent Explainability: Provides token-level contributions, allowing users to understand and trust predictions made by the AI.
• Interactive Dashboard: Facilitates scoring and review processes, while integrating AI-assisted analysis for improved decision-making.
• Production-Ready Packaging: Features seamless model deployment through ONNX, MLflow, FastAPI, and Docker, simplifying integration into existing workflows.
• Optuna Tuning: Incorporates Bayesian hyperparameter optimization to enhance model performance and adaptability.