Experience

Parabol

• United States
• Software Development
• 1 - 100 Employee

• Senior Developer

  • Jun 2020 - Present

  Anyhere in the world Building the future of work



Amazon

• United States
• Software Development
• 700 & Above Employee

• SDE (Multi-Channel Ads)

  • May 2016 - May 2020

  Greater Seattle Area * Single handedly built an end-to-end ML pipeline to predict CTR for Amazon.com's Device Ads. Predictions are used for ad rankings and see ~20% improvement * Designed and built the next generation of ad delivery system to handle tens of millions of semi-connected Kindle devices * Helped to build high performance distributed system to support Amazon.com's Multi-Channel Advertising Server. Strict SLA (<100 ms) and several TB data per day

• SDE (Prime Now)

  • Apr 2015 - May 2016

  Greater Seattle Area * Designed and built the backend service powering the search (~60% of the total traffic) of PrimeNow.com: Able to handle up to 1,500 TPS with average latency of 600ms * Helped launch of PrimeNow.com/search in 6 months, from the team of two engineers; Troubleshooted and resolved hundreds of launch blockers and issues * Designed and developed the asynchronous ImageView for PrimeNow iOS app, reducing latency to half; Added features to support iOS Italy launch, drawing more than 70K… Show more * Designed and built the backend service powering the search (~60% of the total traffic) of PrimeNow.com: Able to handle up to 1,500 TPS with average latency of 600ms * Helped launch of PrimeNow.com/search in 6 months, from the team of two engineers; Troubleshooted and resolved hundreds of launch blockers and issues * Designed and developed the asynchronous ImageView for PrimeNow iOS app, reducing latency to half; Added features to support iOS Italy launch, drawing more than 70K downloads in the first month

• SDE (Kindle Ads)

  • Jan 2013 - Apr 2015

  Greater Seattle Area Work at Kindle Ad Product Optimization team. Projects include: Real-time ad impression prediction and filtering system * Improved prediction accuracy from 248% to 113% * Reduced over-delivery from 232% to 111% * Reduced the duration of daily ad pushing job from 7+ hours to 4 hours Metrics collection data pipeline * Hourly EMR job to process Kindle logs; throughput at 11GB per hour * Automated metrics attributions for data reporting and anomaly detection *… Show more Work at Kindle Ad Product Optimization team. Projects include: Real-time ad impression prediction and filtering system * Improved prediction accuracy from 248% to 113% * Reduced over-delivery from 232% to 111% * Reduced the duration of daily ad pushing job from 7+ hours to 4 hours Metrics collection data pipeline * Hourly EMR job to process Kindle logs; throughput at 11GB per hour * Automated metrics attributions for data reporting and anomaly detection * Reduced delivery feeback loop from 3 days to 3 hours Ad bucketing system * Developed system to allocate different tiers of ads to devices so that customers can get various kinds of ads * Improved ad mix rate by 34% * Boosted CTR from 0.98% to 1.12% Internal website for metrics surfacing and data visualization * Developed MySQL backed website using Django framework * 158 tables, 85.13GB of data, 20+ hourly data loading and reporting jobs running 24/7 * NVD3 graphs; support for real-time ad delivery wrangling



Amazon

• United States
• Software Development
• 700 & Above Employee

• Software Development Engineer Intern

  • Jun 2012 - Aug 2012

  Greater Seattle Area Worked at Kindle ad team. Designed and developed a testing framework for our impression prediction system. Able to reduce the latency of feedback loop from days to hours.



Georgia Institute of Technology

• United States
• Higher Education
• 700 & Above Employee

• Research Assistant

  • Aug 2009 - May 2010

  Greater Atlanta Area * Wrote a simple Pac-Man game using Java to support the research experiments * Demonstrated that learning agent who is able to interact with the 'teacher' only needs 80% demonstrations to outperforms the self-learning agent, by 19%. * Second author of an accepted paper at IEEE conference