Data Scientist & AI Engineer, Nexus
Shipped: IEEE-CIS Fraud Risk Analytics
End-to-end fraud analytics on the IEEE-CIS dataset (590k transactions): DuckDB staging, BigQuery/dbt models with tests, and a LightGBM model evaluated on a time-aware split (ROC-AUC 0.913), served through a FastAPI dashboard. The write-up covers where the model fails, not only the headline score.
Fraud DetectionData EngineeringMachine Learning






Comments
0 comments