Answers are grounded in this paper and cite their sources.
Manual resume screening and single-purpose recruitment tools leave a persistent gap: no existing platform combines candidate-facing career support with recruiter-facing evaluation tools while keeping the scoring process transparent and auditable. This paper presents SmartCareer, an integrated AI-powered career intelligence platform that unifies eight candidate- and recruiter-facing modules—domain-aware resume analysis, skill prediction, applicant tracking system (ATS) scoring, resume enhancement, course/certification recommendation, a voice-enabled mock interview simulator, and recruiter-side resume–job matching with shortlist lifecycle tracking—within a single integrated system. Unlike prior work, which typically addresses one stage of the hiring pipeline in isolation, SmartCareer links resume analysis directly to interview preparation by generating interview questions from the candidate’s own resume and domain profile, and it replaces opaque, black-box matching scores with an interpretable, weighted multi-factor composite score that both candidates and recruiters can inspect component by component. The platform is built as a React.js single-page application with a Flask RESTful backend organised into blueprints, and a hybrid MongoDB/MySQL persistence layer. Machine learning components were trained on 10,000 synthetic resumes (5000 IT, 5000 non-IT): a Random Forest model performs multi-label skill prediction, Logistic Regression performs job-role classification, and a MiniLM-L6-v2 sentence-transformer computes resume–job semantic similarity. The mock interview module evaluates spoken responses using Faster-Whisper transcription and Librosa-derived acoustic features—pitch variability, RMS energy stability, pause ratio, and filler-word frequency—to generate feedback on confidence, clarity, delivery, and articulation. On a held-out test set, the system achieved a macro-F1 of 1.00 for IT skill prediction, 100% accuracy for IT job-role classification, 91% accuracy for non-IT job-role classification, and approximately 17% top-class accuracy for course/certification recommendation, a result discussed critically in light of the large label space involved. Beyond these benchmark figures, we report a component ablation showing that the highest-weighted matching component degrades rather than improves ranking quality, a learning-to-rank derivation of the scoring weights, an empirical calibration of the recruiter ranking thresholds that finds them to carry little information about role fit, an end-to-end API benchmark confirming that throughput does not scale with concurrency in the current single-process deployment, and a noise-robust replacement for the interview audio feature pipeline validated by controlled noise injection. Evaluation against two external corpora of independently annotated real resumes shows that the parser’s fixed skill vocabulary can represent only 20.8% of the skills human annotators marked, and nothing at all for 41.0% of candidates, so the synthetic benchmarks above substantially overstate readiness for real input. The contribution is therefore architectural rather than predictive: SmartCareer demonstrates that candidate-facing and recruiter-facing career tooling can be built as one connected pipeline with a fully auditable, component-level scoring model, and it reports, rather than conceals, where that pipeline currently fails.
No takes yet. Share an insight, caveat, or question.
Rane et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: