Bilingual extraction
Arabic and English CVs and job descriptions parsed and normalized — skills
Government · Qatar
AI job-matching engine
An explainable hybrid matching stack serving a national employment platform — semantic matching at scale with a score breakdown both applicants and recruiters can read.
88% top-3 match precision vs the legacy rule engine
7,000+ public and private sector users
<3 s to a ranked recommendation list — down from 90
65% less manual CV triage
The challenge
What was in the way.
- More than 60,000 active jobseekers outpaced any manual résumé screening, delaying hiring and nationalization targets.
- Rule-based filters on education and eligibility missed nuanced skill fit and produced low match precision.
- Applicants got no explanation for rejections, eroding trust and repeat engagement.
What we built
The system, in brief.
Mandatory-rule gate
Citizenship
Semantic matcher
A dual-encoder model fine-tuned on local corpora computes similarity
Explanation chatbot
Score breakdowns — skill overlap
Quarterly learning
Hiring outcomes feed retraining each quarter
Outcomes
What changed.
- 88% top-3 match precision in A/B testing against the legacy rule engine, across 15,000 vacancies.
- Manual CV triage hours fell 65% — roughly five recruiter FTEs re-allocated.
- Recommendation lists generated in under 3 seconds instead of 90.
- Transparent explanations lifted applicant satisfaction 18%; all model decisions logged for seven years.
Client referenced by sector and country · detailed references on request