Led a team of 5 engineers to deliver the Ad Content Relevance feature end-to-end, designing backend control and data planes, coordinating front-end updates, and overseeing asynchronous ad attribution pipelines; mentored junior and mid-level engineers through pair programming, design reviews, and code reviews.

Drove cross-org collaboration and implementation for a senior vice president-level initiative to optimize Amazon DSP’s (Demand Side Platform) predictive ML models (probability of app download, click, purchase) by 8%. Partnered with Applied Science and DSP engineering teams on feature engineering, model training/evaluation, and online inference.

Designed and implemented an end-to-end ML pipeline for ad recommendations, including raw data discovery, automated data ingestion, and deployment using AWS Glue, Apache Spark, Airflow, and batch inference.

Hosted biweekly office hours to support Device Ads publisher teams, serving as the integration point for metadata retrieval requests and helping 10+ partner teams successfully integrate with or request features on our service.

Launched a closed beta of a content relevance optimization feature for Prime Video display ads by implementing service-level integrations with our ad personalization system and conducting A/B testing to gradually roll out the feature to users.

Collaborated across teams to deliver new features for Fire Tablet and Alexa placements, including enabling early filtering of ads based on content restrictions and adding enrichment changes to support new placement launches (e.g. Fire Tablet display ads, Alexa skill ads)

Built high-availability (99.5%) Java/Spring Boot ML model hosting infrastructure delivering real time fraud predictions, preventing $60M in potential monthly losses.

Implemented a Tier 0 risk evaluation checkpoint using rules and distributed caching (Memcached) to achieve <10 ms latency for real-time risk assessment.

Engineered horizontally scalable data and model-serving pipelines, increasing throughput 10× to support more complex models and higher inference volumes.