@ Foundation Data - Data Storage

@AWS Amazon Fraud Detector - AutoML

● Joined team when service in private Beta, worked with team to make it launch in 2019 re:Invent.

● Leading S-Goal project: launching Account-Take-Over model type in 7 regions. Re-architected container-based enrichment system into SDK based sub-system; Re-architected aggregation during inference from Apache Flink based into Lambda based data ingestion system.

● Leading S-Goal project: launching Transaction-Fraud-Insignt model type in 7 regions. Designed and implemented Data Lake and query engine to generate model training raw data; Lead and passed AppSec review; Owned SDK release process.

● Tech lead to make Amazon Fraud Detector service PCI-ISO compliance.

● Tech lead to design and implement Scientist Experimentation Platform to support troubleshooting customer facing issue without eyes-on customer data.

@Amazon Prime - AutoTargeting

● Designed and implemented offline data processing pipeline using PySpark with ability to process 200MM+ records, and generated segment level model performance results used inside WBR report.

● Built automation pipeline for data loading and reporting jobs, set up Tableau WBR report to review clients' model performance.

● Built rule-based automating selection tool to reset WW 200+ marketers' experiment baseline so that all experiments have incremental performance.

● Implemented offline simulation tool and reduced 80% time to evaluate performance of different feature generator methods (Raw, Prime2Vec, XGBoost) combined with several scoring algorithms (BLIP, BLR).