AI-Driven Career Guidance for Vocational Students — A Controlled Trial in South Kalimantan, Indonesia

Researchers based at Universitas Negeri Yogyakarta's graduate school designed and tested an AI-driven career-guidance system for vocational (SMK) students, combining three linked components: a supervised machine-learning model that maps a student's skills against labour-market demand; an adaptive mentoring module pairing students with an AI chatbot and human career counsellors and industry mentors; and a natural-language-processing module that analyses job postings to surface real-time labour-market intelligence.

The system was piloted with 320 students recruited across three vocational schools in South Kalimantan province — SMK Isfi Banjarmasin, SMK Telkom Banjarbaru and SMK NU Banjarmasin — with 15 career counsellors and 12 industry mentors also involved. The core evaluation compared 90 students who used the AI system for 12 weeks against 90 matched control-group students who did not, in a mixed-methods design; the peer-reviewed write-up reports on this 180-student intervention/control comparison. The skills-mapping model achieved 87% accuracy (precision 0.85, recall 0.83, F1 0.84, ROC-AUC 0.89) in predicting suitable career pathways.

Students' self-reported career-path anxiety fell from a mean of 73.1 to 53.6 in the intervention group (p<0.001), a 26.7% reduction, versus a much smaller change in the control group. Engagement was substantial: 65% of intervention students repeated a skills-gap analysis three or more times, 79% took part in AI-mediated mentoring (2,400+ chatbot interactions logged), and average satisfaction with the mentoring component was 4.60/5.0. The system's labour-market module analysed 500 job postings. The study was published in F1000Research (indexed on PubMed) in February 2026, after an earlier preprint in October 2025.

The authors themselves flag several limitations: the evaluation covered a single province and a modest sample over only 12 weeks, so long-term effects on actual employment outcomes remain unexamined; the sample may not represent Indonesia's regional diversity; and they explicitly note a risk of algorithmic bias from the historical labour-market data feeding the system. The study was also run by the tool's own developers rather than an independent evaluator.

Read the full analysis: https://pubmed.ncbi.nlm.nih.gov/41869558/