At Pinterest, I modernized advertiser analytics processing petabytes of data daily, cutting P90 latency by 50%, reducing serving instances by 68%, and delivering roughly 10-second freshness. I built batch and streaming pipelines supporting 200K+ attributed events per second across Ads Manager, billing, pacing, and reporting.
I also drove key parts of Pinterest's Druid-to-StarRocks migration and optimized Spark and OLAP workloads, reducing a 10+ hour preprocessing workflow to about 15 minutes. My work spans governed analytics and ML datasets, ML infrastructure, architecture, reliability, capacity planning, and mentoring engineers.
Previously, I built real-time fraud and analytics pipelines at DiDi and reusable ML data platform components at Google, including TFX and Apache Beam capabilities for production models at scale. I turn raw data into reliable products that help teams make better decisions and scale with confidence.
