Data Engineering

Monetization / Revenue Platform

Incentive Allocation Platform - how to optimally allocate incentives across the world, to balance future supply and demand in Uber's marketplace. Involved data engineering, machine learning, and optimization.

Tech: Spark, Hive, Presto, Python DS stack (pandas, NumPy, etc.)

Architect for ebay.com next-gen experience's Search, Browse and Listing platforms (Backend).

I joined an organization that was started with the goal of modernizing eBay's tech stack. We adopted state-of-the-art ML technologies from across the industry and applied them in e-commerce space to improve online shopping experience.

• Cloud based stack leveraging many open source technologies: TensorFlow, Spark, Docker, Kubernetes, Play Framework (Scala), etc.

• Streaming architecture with near real time pipelines keeping the Search index up-to-date.

• Selection and ranking powered by deep learning models: semantic vectors for relevance, category prediction and NER for filtering and ranking, click prediction for re-ranking, etc.

• The new platform, built from scratch in 2 years by 60 engineers, is 15x faster and resulted in 7x conversions compared to old stack.

• Along with traditional search experience, the platform also supports eBay's chatbots on FB Messenger (eBay ShopBot) and Google Assistant.

As member of Architecture team, I led high-impact broad initiatives in a number of key product areas.

Windows

• Ranking service to replace the rules engine in Windows Engagement and Monetization System.

AOL Transition

• Led the creative migration track as part of transitioning Microsoft’s display ads to AOL.

MSN

As part of migrating MSN website to a new platform,

• Implemented ad size prediction algorithm that helped prevent 900 potential incidents by detecting ad size assignment mistakes. In addition, it was used for pre-allocating correct sized ad containers on MSN, improving user experience.

• Worked on a new BI product that reduced monitoring cadence from daily to hourly during MSN platform migration.

• I also worked on MSN ad performance improvement initiative on the new platform.

I played a critical role in architectural redesign of Bing's Product Ads.

Bing Product Ads

• Designed and built Selection and Ranking Server, a high performance c++ distributed server with load balancing, configurable protocols, watchdog, built-in logging, monitoring and metrics, test client, unit and component testing frameworks, scripts to aid deployment, etc.

• Pipeline to ensure data quality of Product Ads. With 98% accuracy, the Judgment Engine algorithm I created dramatically reduced Bounce Rate / Quick Back Rate on Bing, improving user experience

• Performed analysis on Product Ads Ad Selection, to find opportunities to improve the selection models.

Hackathons

• Demo'ed custom alerting features on Cortana. This hack won us first prize in the hackathon. These features have been adopted by Cortana team and were launched in production.

• Demo'ed location based answering system on Cortana leveraging Product Ads. This hack won us third prize in the hackathon.