Apr 2021 – Jan 2025 · 3 yrs 9 mos · Palo Alto, California, United States

• Founded a new team and grew the team to 12 people within 2 years. My team supports end-to-end inventory forecasting solutions, delivery planning and contention simulation for all Amazon guaranteed ad products. All products my team owns operate on the extremely large data set (raw bid logs)
• Owns delivery planner which optimizes billions of dollars of budget allocation and budget distribution between publishers for all guaranteed ads in Amazon across all media formats (video, audio, display, devices) using Linear Programming to maximize revenue, and ensure delivery rate and smooth pacing. Despite the complexity of the computation, the proprietary LP solver the team builds is capable of completing 10,000 rounds of projected gradient descent to find the optimal budget allocation of thousands of ads and tens of millions of users within 10 mins by parallelizing computations with 8 GPU cores
• Expanded the scope of the team by proposing and developing ads delivery simulators that supported ads supply forecasting and planning. The simulator processes 300,000 simulations daily, forecasting over $60MM in daily revenue. In addition, the simulator has been widely adopted in producing critical business reports (e.g. Sell-through rate), and guiding Sales team to pitch over $800MM of STV upfront commitments in 2023

Sep 2018 – Apr 2021 · 2 yrs 7 mos · Palo Alto

• Designed and implemented delivery planner which improved video ads delivery in full rate from 70+% to 98%
• Proposed and built end-to-end forecasting solution for Amazon device ads

Mar 2016 – Aug 2018 · 2 yrs 5 mos · San Jose

• Co-developed the team's first machine learning service pipeline and framework, and launched the price prediction service
• Designed and implemented price outlier detection service
• Developed a large-scale Kafka index consumer that processed inventory metrics and indexed results into ElasticSearch

Feb 2015 – Mar 2016 · 1 yr 1 mo

• Designed and implemented a Gradient Boosting Machine model to predict avoidable truck rolls
• Developed archiving functionality for Kafka consumer and built batch and realtime Kafka data ingestion pipelines

Jul 2013 – Feb 2015 · 1 yr 7 mos

• Contributed to the launch of oxygen.com website
• Contributed to open-source Drupal codebase
• Led backend development including content rendering, URL validation and generation and revision scheduling

May 2012 – Aug 2012 · 3 mos · San Francisco Bay Area

• Developed custom MapReduce jobs in Python for ads performance analysis
• Proposed improvements to the ad targeting algorithm

Jan 2012 – May 2013 · 1 yr 4 mos · Greater Philadelphia Area

• Project: Clinical Gait Analysis with Kinect
• Designed and implemented a real-time human tracking and following system on a mobile robot
• Designed and implemented a Hidden Markov Model (HMM) human motion tracking model to classify human gait cycle
• Project: One-shot-learning Gesture Recognition
• Designed and implemented a maximum likelihood path HMM model to estimate gestures with only one training example.
• Won second place of ChaLearn Gesture Recognition Challenge and the work was presented in CVPR '12 workshop

Jun 2010 – Jun 2011 · 1 yr · Hong Kong

• Project: Processing Private Quires over Untrusted Data Cloud through Privacy Homomorphism
• Accepted and Published in ICDE '11
• Initiated one optimization method and improved the others
• Implemented the entire framework with Java and conducted performance analysis
• Project: iGreen: Enabling Privacy Preserving Location-based Services (Undergraduate Thesis)
• Selected to be sponsored by Li Ka Shing Foundation
• Bronze Prize in 2011 Amway University IT Project Competition
• Individually designed and implemented the entire system on iPhone and published in Apple App Store (iGreen Hong Kong) (currently offline due to the expiration of developer license)