ML insights, Data visualization, ML infra

▪ Fullstack at local services Ads.

▪ Web app for creating search query understanding models. Reduced model building time from 1 month to 1 week.

▪ Online instant quotes on requested services. Improved advertiser feature adoption by 30%.

▪ Automatic pipeline to collect human evaluations on lead conversion, powering machine learning model training. Improved evaluation efficiency by 50%.

Real time credit card application fraud ring detection based on graph analytics. Explored and exploited every data source available in real time, to bring customers a better experience and to catch fraud timely. Involved in feature pruning (partial dependency plot), feature creation, extendable model pipeline building, and hyper parameter tuning (downhill simplex). Built dashboard to easily track model performance.