Built and led cross-functional teams of 25 people to deliver the first version of internal translation management system with neural machine translation engine. Responsible for customer listening tours, translation memory and infrastructure design, Scrum practice, performance analytics, post-release operation support, and vendor management.* Shortened launch time by a year by initiating buy-vs-build discussion with leadership and convincing them to bootstrap development with purchase of 3rd-party source code license.* Defined, collected, and analyzed 200+ application metrics to lower support cost and discover new use cases. Post-launch hot-fixes were 30% lower than that of other Amazon teams.* Kept team attrition rate 20% below Amazon average by initiating software development learning program for non-technicals, monthly leadership rotation, cross-function mentoring, and continuous process improvement activities identified in sprint retrospective.* Re-energized the team with refresh 1-, 3-, and 5-year roadmap to make localization first-class feature to AWS CodePipeline continuous delivery service.

Led a team of 10 developers to deliver the cloud-based file storage application. Responsible for architecture design, security compliance, Game Days launch readiness, and alarm audit. * Re-architected and migrated the monolithic backend image processing service to more scalable micro-services so that we could accelerate the release to App Store.* Leader of customer voice team to design and implement A/B testing, in-app feedback collection and analysis from millions of customers.*Increased request handling throughput by 100% by using horizontal scaling on the Amazon SQS: increasing number of threads per client and batching multiple messages with a single SendMessageBatch call.

Led a team of 5 developers to customize Office SharePoint for Education. Responsible for requirement collection, user study, and collaboration across different SharePoint teams.* Reduced storage cost for education institutions from 5X to 2X per GB by applying the latest incremental storage backup solution from Windows team.* Reduced JavaScript load time by 30% by collaborating with 30+ internal teams to refactor the libraries.* Built a modular platform and adopted the failed-fast-learn-fast way of experimenting new features to accelerate the iterations for new institutions.

* Designed and implemented the customization of generic machine translation output without the need to retrain and maintain translation model.* Improved the fuzzy match of segments by 25% by using sub-segment match and custom metadata such as usage and recency data collected from customers.* Speed up the indexing of translation memory by 300% by both horizontal and vertical scaling and the use of distributed job scheduler.