ML-Based Candidate Evaluation with Automated CV Extraction

Abstract

Recruitment often relies on human evaluation of CVs, which can be time-consuming and subjective. This study presents a hybrid approach to streamline candidate assessment by combining automated CV extraction with machine learning. We use Google Gemini API to extract structured data from PDF CVs, including skills, education, experience, and previous roles. Since AI-based initial scoring can be biased, we train a machine learning model, specifically a Passive-Aggressive Classifier, on these features to predict candidate levels (Junior, Mid, Senior) consistently. Our approach ensures unbiased and reproducible evaluation, demonstrating that while automated extraction accelerates data processing, machine learning provides accurate and fair candidate classification. Results show that similar CVs are consistently categorized by the model, overcoming inconsistencies observed in initial AI ratings.