Public employment programmes need timely evidence on occupational gaps by department, but GEIH microdata cannot be shared openly. Our group harmonised GEIH 2022–2025 into a partitioned Parquet lakehouse, documented lineage with audit logs and a schema contract, and orchestrated a Prefect ETL flow to a curated monthly aggregate by department. We trained linear and small MLP models on official aggregates with train/validation/test splits and built a dashboard separating official GEIH indicators from optional news context. Differential privacy and k-anonymity rules were applied before publishing any table. Results show persistent regional disparities in employment rates; the linear model was preferred for decision audiences while the MLP highlighted non-linear patterns for analysts. We recommend targeting two departments for pilot interventions and publishing only aggregate indicators with documented governance controls.