FinTech · Real-Time Anti-Fraud ML Platform
Role: Senior Data Scientist / ML EngineerFinTech · Fraud Detection · Real-Time InferencePrior experience — Michele Sanna at RevolutStatus: Completed
Designed and shipped a two-stage detection pipeline, sub-40ms gradient-boosted scoring at the transaction layer, plus a GraphSAGE GNN over account/device/IBAN graphs to surface mule networks and coordinated fraud rings.
- Python
- PyTorch
- DGL
- XGBoost
- Kafka
- Flink
- Feast
- Redis
- Kubernetes
- MLflow
Scroll to explore →
−47%false positive rate on card-not-present transactions within 6 months
+31%fraud capture rate (value-weighted) vs. legacy rules engine
~€18M/yearestimated fraud losses prevented (cards + SEPA combined)
p99 220ms → 38msinference latency reduction
Next project
J.P. MorganNext
Want this kind of work in your business?
Start with the problem, not the platform. Tell us what's slow, manual or breaking.