Kevin Shi

Led end-to-end development of Netflix’s first large-scale ads ranking platform, improving ad CTR 8–10% and fill rate 5%. Architected real-time multi-task inference and a budget pacing/campaign allocation engine, increasing budget delivery accuracy to 97–99% while reducing spend variance by 30%.

Built personalization and recommendation systems for Amazon Personalize, designing ML recipes and large-scale ranking pipelines that improved recommendation coverage by up to 1.8×. Integrated LLM-powered recommendation content workflows and automated end-to-end MLOps retraining/deployment to improve model reliability and reduce retraining overhead.

Led development of real-time CTR/CVR prediction and auction optimization pipelines for Twitter Revenue Science, improving ad ranking efficiency and advertiser targeting quality. Built low-latency online inference and streaming feature engineering systems to reduce serving latency and improve prediction accuracy during live events.

Led machine learning initiatives for sponsored search and advertising systems by building CTR/CVR prediction, ranking optimization, and large-scale behavioral feature engineering pipelines. Developed low-latency ad ranking and inference pipelines with caching and auction scoring, supporting large search traffic under strict SLAs and enabling safe staged rollouts via online A/B experiments.

Built CNN- and BERT-based listing understanding and attribute extraction systems, improving listing metadata quality and search relevance. Developed floor plan computer vision models and scalable geospatial enrichment pipelines (S2/geohash) to improve categorization and location intelligence for millions of listings.

Built classification models for real-time detection of fraudulent listings, including a contextual multi-armed bandit approach created from scratch. Created feature pipelines, monitored online performance, and partnered with operations to design review workflows and metrics for queued review efficiency and efficacy.

Earned a Master of Science in Computer Science from the University of California, Berkeley between 2016 and 2018.