Decision systems for consequential environments

In a regulated BNPL environment operating in Germany and Austria, I led data-science work across fraud prevention, payment risk, consumer-risk decisioning, transaction monitoring and portfolio oversight for more than €200 million in annual transaction volume.

I developed fraud and risk models combining transaction behaviour with privacy-sensitive device, identity and location intelligence, alongside credit information and relevant external data. These models supported decisions about approval, review, monitoring and intervention.

I built an in-house fraud capability that replaced reliance on an external provider. It connected model outputs with analyst review, case management, intervention rules, feedback loops and ongoing monitoring.

I developed an internal orchestration and decision-intelligence layer integrating machine-learning models, rule engines, generative AI and retrieval-augmented internal knowledge. It brought model outputs, automated checks, risk alerts and relevant internal guidance into shared workflows for human review.

One merchant-assessment workflow combined tabular risk classification with website-content analysis using a compact language model. It helped identify inconsistencies between assigned merchant-category codes and actual business activity.

Retrieval-augmented generation provided access to internal policy and operational knowledge. Source traceability, escalation paths and analyst review connected that information to operational decisions.

I developed monitoring and executive-reporting frameworks covering fraud exposure, bad debt, portfolio performance, model behaviour, operational effectiveness and emerging risk patterns.

The work connected data science with product, engineering, risk, operations, compliance, legal and executive stakeholders. Model documentation, reproducibility, explainability and human oversight supported operational governance and audit readiness.