The Challenge
Crestwillow Financial is the client behind this project. A Dallas financial services business processed customer transactions and related payments. It needed to identify suspicious activity while making the investigation process more manageable.
High transaction volumes and changing fraud patterns complicated detection. Too many false positives could also increase the review team’s workload, so the solution needed to balance automated monitoring with useful investigation information.
What ApexCore Digital built
ApexCore Digital implemented a hybrid approach combining rule-based checks and machine-learning anomaly detection. Models trained on historical transaction data analyze incoming streams for suspicious patterns.
Kafka supports streaming, while Python and FastAPI support processing and service access. Risk scores, real-time alerts, customer behaviour analysis and transaction histories feed a React investigation dashboard. Reporting and model monitoring support ongoing oversight of detection activity.
How the workflow supports the business
The investigation workflow starts with transaction information and the signals identified by the detection system. Rules provide explicit checks, while anomaly models examine patterns using historical data. Risk scores, alerts and transaction history then support review through the investigation dashboard. This combination gives analysts context around flagged activity instead of relying only on an alert. Model monitoring and reporting are included because detection requires ongoing oversight as transaction behaviour and fraud patterns change. Automated signals support investigation; they do not establish that every flagged transaction is fraudulent.
Reported Results
- 40% reduction in manual transaction review.
- 30% reduction in false positives.
- Real-time transaction monitoring introduced.
Technology: Python, Scikit-learn / TensorFlow, FastAPI, PostgreSQL, Kafka, React, model monitoring, AWS.
