Kevin Shi

Led end-to-end development of Netflix's ads ranking model, evolving a GBDT baseline into a multi-task neural network. Designed and productionized a budget-pacing and campaign-allocation engine, and architected a streaming ML data platform.

Led recommendation-model development and evaluation for Amazon Personalize, covering recipe design, metadata-aware features, and hyperparameter optimization. Shipped Popularity Tuning and designed the LLM pipeline behind Content Generator.

Tech-led the Ads Prediction stack supporting Twitter's ad-serving pipeline, including real-time CTR, conversion, engagement, and advertiser-value models. Reduced serving latency and built streaming feature pipelines.

Developed CTR and conversion prediction models for Amazon Sponsored Products across search-result and product-detail pages. Improved tail-query and new-product relevance using semantic similarity and cold-start strategies.

Built CNN-based image classifiers to identify floor-plan images in listing photos and fine-tuned BERT models to extract structured attributes from host-written content. Developed geospatial enrichment pipelines using Scala, Spark, Google S2, and Geohash.

Master of Science in Computer Science from the University of California, Berkeley, from 2016 to 2018.

Doctor of Philosophy in Operations Research from the University of California, Berkeley, from 2013 to 2018.