Real-time fraud detection for a FinTech platform

Problem

Manual review of transactions caused delays and missed 23% of fraudulent activity.

Solution

Built an ML-driven scoring engine processing 50K+ transactions/sec with sub-100ms latency.

Approach

Model evaluation against labeled outcomes, shadow mode rollout, and operational dashboards for reviewers with clear override and audit trails.

Tech stack

• Python
• Kubernetes
• PostgreSQL
• Google Cloud
• PyTorch

Results

• 94% fraud detection accuracy
• 67% reduction in manual reviews

Related capabilities

• AI & intelligence
• Application & platform engineering
• Security & QA