At Media.net, I designed and shipped three URL-derived features for the conversion-prediction model, delivering a 29.7% revenue increase, an 18% conversion-rate increase, and a 7% lower cost per acquisition in a production A/B test. I also took a partition-routing system to 100% production traffic across two ad surfaces.
I engineered distributed feature-serving paths and built tools that made experimentation faster and more accessible, including a Kubeflow pipeline that cut the build-test-deploy cycle from three days to one. I also designed and shipped a production bid-calibration system, and built an in-app data visualizer that reduced query latency from 40 seconds to five.
