Sara Fibla Salgado

Designed and built an end-to-end transaction-level fraud detection framework using ML, addressing low precision in existing merchant-level alert systems across millions of daily transactions. Engineered the full ML pipeline: data integration, feature engineering, label design, leakage controls, model training, imbalance mitigation (SMOTE/undersampling), threshold optimization, and interpretability (SHAP). Architected the project across three phases: Data Engineering, Model Development, and Business Impact; collaborating with stakeholders to translate model outputs into actionable production decisions.

Engineered Python and Java automation tools for inventory tracking and order processing, reducing manual reporting workload by 8 hours/week and improving operational accuracy across 3 departments. Collaborated with cross-functional teams to identify bottlenecks and implement scalable digital solutions, redesigning manual workflows to enhance productivity and reduce error rates.