I care as much about the customers a system wrongly accuses as the ones it rightly catches.

German-Jordanian, Berlin-based, working in Arabic, German, and English. Six years across four sectors — cloud, fashion, automotive, energy — taught me the same lesson in four dialects: the expensive errors in abuse prevention are rarely the cases you miss. They're the legitimate people you break in the process, and the trust you spend doing it.

So I work the precision/recall trade-off as the actual job, not a footnote to it. That means treating a flag as a hypothesis to disprove, building carve-out gates before bulk actions, refusing to ship a recall gain that's quietly paid for in false positives I'm not looking at, and measuring whether a control worked with a holdout rather than a flattering before/after. I have a genuine streak of epistemic skepticism — I'd rather know why a number is true than be reassured that it is — and in this work that's a feature, not a quirk.

At AWS I build the infrastructure, the methodology, and the deliverables: a data-access layer the investigations run on, LLM agents that argue their way to defensible answers, a whole-population relations graph that makes rings visible, and root-cause work that reads production code to settle a question rather than guessing at it. At Zalando I owned fraud metrics for business reviews and built the Remaining Fraud Damage measure across six European markets. The toolkit is ordinary; the discipline is the point.