Implementation of HR Candidate Recommendation System
HR recommendations work both ways: system matches candidates to vacancy and vacancies to candidate. Complexity: matching unstructured data (resume, JD), soft skills accounting, diversity requirements and compliance constraints. Plus cannot use age, gender, nationality as features.
Semantic Matching
Semantic matching with sentence-transformers gives 20-30% higher accuracy than keyword matching. Screening time for 100 resumes reduces from 3-4 hours to 15-20 minutes. Important: all features must be skills/experience based, without demographic data. Regular bias audit recommended (disparate impact analysis).