BRIDGING EDUCATION AND THE LABOR MARKET: AN EDGE-AI ARCHITECTURE FOR AUTOMATED STUDENT-TO-CAREER MATCHING
Keywords:
Edge Computing, Automated Career Matching, Large Language Models (LLMs), Data Sovereignty, Educational TechnologyAbstract
In the contemporary educational landscape, aligning academic curricula with rapid labor market transformations remains a critical challenge. For academic institutions and career advisors, manually bridging the gap between student competencies and industry demands is unscalable due to the highly unstructured, heterogeneous nature of student resumes (CVs) and corporate job postings. Furthermore, while modern Artificial Intelligence (AI) offers robust parsing capabilities, utilizing commercial, cloud-based Large Language Models (LLMs) to process sensitive student data introduces severe privacy, compliance, and data sovereignty risks. The primary purpose of this study is to overcome these limitations by engineering a secure, automated tool that empowers the "teacher of the future" to align student profiles with industry needs without compromising data integrity. The methodology proposes a privacy-preserving Edge-AI architecture designed for automated student-to-career matching. It utilizes a multi-stage pipeline of decentralized, locally hosted LLMs to perform intelligent data extraction. Specifically, the system employs Qwen 3.5:35B to digitize visually complex job offers via Optical Character Recognition (OCR), Gemma 3:27B for semantic structuring and strict JSON schema enforcement, and TranslateGemma:27B for high-fidelity localization into the Macedonian (or other local) language. A student-centric matching algorithm subsequently evaluates these profiles, computing semantic similarity scores across re-calibrated domains: mandatory prerequisites, competencies, formal education, practical exposure, and logistical alignment. Experimental results validate the efficacy of this local architecture. Executed on edge hardware (NVIDIA RTX 5090), the pipeline achieved a 99.4% JSON schema validity rate for CV structuring and a 100% success rate in deterministic matching outputs, with end-to-end processing averaging under 75 seconds, of which 30 seconds for CV-job matching. The conclusions drawn indicate that localized edge-computing architectures can achieve enterprise-grade parsing and matching while keeping all sensitive data strictly air-gapped, neutralizing cloud-based privacy risks and high costs. Based on these findings, it is highly recommended that academic IT departments and career centers adopt on-premises AI inference to maintain General Data Protection Regulation (GDPR) compliance while modernizing advisory services. Additionally, the data suggests that dynamically adjusting algorithmic weights toward foundational academic knowledge rather than longitudinal corporate tenure significantly improves matching fairness for entry-level candidates, paving the way for future integrations directly into university Learning Management Systems (LMS).
References
Bogen, M., & Rieke, A. (2018). Help wanted: An examination of hiring algorithms, equity, and bias. Upturn.
Fazel-Zarandi, M., & Fox, M. S. (2009). Semantic matchmaking for job recruitment: An ontology-based hybrid approach. CEUR Workshop Proceedings, 525.
Feretzakis, G., Vagena, E., Kalodanis, K., Peristera, P., Kalles, D., & Anastasiou, A. (2025). GDPR and Large Language Models: Technical and Legal Obstacles. Future Internet, 17(4), 151.
Holley, R. (2009). How good can it get? Analysing and improving OCR accuracy in large scale historic newspaper digitisation programs. D-Lib Magazine, 15(3/4).
Kakasevski, G., & Mishev, A. (2017). Optimization and scheduling algorithm for data intensive workflows in distributed data mining architecture. Eurocon, Ohrid, North Macedonia.
Khaouja, I., Kassou, I., & Ghogho, M. (2021). A Survey on Skill Identification From Online Job Ads. IEEE Access, 1-1. 10.1109/ACCESS.2021.3106120.
Ling, Z., Zhang, H., Cui, J., Wu, Z., Sun, X., Li, G., & He, X. (2025). Beyond Human Labels: A Multi-Linguistic Auto-Generated Benchmark for Evaluating Large Language Models on Resume Parsing. 31907-31933. 10.18653/v1/2025.emnlp-main.1626.
Manish, V., Manchala, Y., Vijayalata, Y., Chopra, S.B., & Reddy, K.Y. (2024). Optimizing Resume Parsing Processes by Leveraging Large Language Models. 2024 IEEE Region 10 Symposium (TENSYMP), 1-5.
Montuschi, P., Gatteschi, V., Lamberti, F., Sanna, A., & Demartini, C. (2014). Job Recruitment and Job Seeking Processes: How Technology Can Help. IT Professional 16(5):41-49.
Spada I, Fabbroni V, Chiarello F, Fantoni G (2024) Standardising job descriptions in the humanitarian supply chain: A text mining approach for recruitment process. PLoS ONE 19(7): e0305961.
Suleman, F. (2018). The employability skills of higher education graduates: Insights into conceptual frameworks and methodological options. Higher Education, 76:263-278.
Zhao, W. X., Zhou, K., Li, J., Tang, T., Wang, X., Hou, Y., ... & Wen, J. R. (2023). A survey of large language models. arXiv preprint arXiv:2303.18223.
