Badges
Certifications
dirk_lennarz has not earned any certificates yet.
Work Experience
Applied Scientist
Amazon•  December 2021 - March 2023
As part of the Buyer Risk Prevention at Amazon, I work to keep Amazon stores safe and trustworthy for customers by ensuring their accounts are not used to commit fraud. My responsibility is to prevent fraudulent orders in the EU region for digital items, leveraging Machine Learning (ML) and business rules to meet various business goals. In buyer seller collusion, fraudsters create fake seller and buyer accounts and leverage e.g. stolen credit cards to buy items from themselves to cash out on royalties. To accomplish this, they create fake digital content, like gibberish eBooks or fake apps. For fake eBooks, I developed novel business rules to stop this kind of fraudulent orders and provided data tags that served as input for a fake eBook detection model. I stopped the fake app attack with novel business rules, but also created a ML model leveraging semantic product embeddings to detect fake apps at upload time. One of the main business goals is the dollar amount of chargeback received from banks, e.g. when customers dispute charges on their stolen credit cards. During my tenure, I have reduced that number by 60% for a major EU country, whilst at the same time reducing the number of business rules that need to be maintained and updated by 12%. ML models in the fraud space suffer from performance deterioration over time due to changing attack patterns. We regularly retrain models to keep them performant, but most business rules use thresholds on model scores and retraining can have a significant impact on the volume, potentially putting business goals at risk. This is why we perform a scoreshift calibration after every model refresh to keep the rule volume constant. I have developed an end-to-end pipeline in Python that automates this process, reducing the time it takes scientist to accomplish this task from days to hours.
Education
Ruprecht-Karls-Universität Heidelberg
Physics, PhD•  September 2009 - October 2012
PhD Thesis: Transient Very-High-Energy Gamma-Ray Emission from Gamma-Ray Bursts and Supernovae with H.E.S.S.
Links
Skills
dirk_lennarz has not updated skills details yet.