Manual resume screening does not scale to the volume of applications a single vacancy now attracts, while conventional Applicant Tracking Systems rely on rigid keyword matching, evaluate one job domain at a time and offer no explanation for their decisions. This study designed, implemented and evaluated a machine learning model for bulk resume classification and recommendation, integrating a fine-tuned BERT classifier, LLaMA2- and Flan-T5-based recommendation generation and a LIME explainability layer within an ethical governance framework. A Design Science Research methodology, combined with Design Thinking and an Agile life cycle, was executed across six two-week sprints. The dataset comprised 14,450 labelled resumes spanning five job domains, split 70/15/15. The fine-tuned classifier achieved 91.3% accuracy and a macro-averaged F1-score of 0.893 on 2,168 held-out resumes, outperforming Support Vector Machine, Random Forest and Naïve Bayes baselines by 8.8 to 16.5 percentage points. A Demographic Parity Ratio audit across gender returned 0.96–1.04 in every domain. User Acceptance Testing with 30 Human Resource professionals in Nakuru County, Kenya returned Technology Acceptance Model scores of 4.30, 4.17 and 4.23 for perceived usefulness, ease of use and fairness, each above the 4.0 adoption-intent threshold.