Abstract:
This paper documents the methodological evolution of job classification within the ECES Labor Demand project, charting a deliberate progression from manual methods to a sophisticated AI system. The culmination of this work is JobIt-CLF, a novel system built on an agentic architecture that combines a Large Language Model (LLM) with a specialized SQL Agent for granular, 4-digit ISCO-08 coding. By grounding its reasoning in a comprehensive database of official ILO definitions, the system mimics expert human logic through a transparent, hierarchical process. Rigorously validated against both large-scale automated audits and expert human review, JobIt-CLF achieves 94–97% accuracy. By detailing this multi-stage evolution, we present a robust, replicable framework for aligning large-scale labor market data with international standards, ensuring more impactful economic analysis.
