Lailin Chen

My focus is deep learning for computer vision and NLP, both research and engineering. I drive the adoption of recent state-of-the-art research into many aspects of Amazon business. I also mentor multiple hardware-software co-optimization projects inside the company. I have created multiple open research projects for and supervised multiple researchers in UC Berkeley. I built the computer vision group in CoreAI Seattle from ground up, starting from being the only person with experience in deep learning and computer vision to a group now works on multiple projects. I trained SDEs to become deep learning practitioners and brought top performing SDEs and industry computer vision experts on board. I have also invented a virtual team collaboration model that allowed us to quickly ship our work into production and scale.

Computer Vision / Machine Learning Team Worked on Face Recognition, Document Detection, and OCR. I implemented and optimized the OCR deep net in Tensorflow in this blog: https://blogs.dropbox.com/tech/2017/04/creating-a-modern-ocr-pipeline-using-computer-vision-and-deep-learning/

Projects (from most recent to least recent): Inventory Planning Research Lab: - Led a team of statisticians and SDEs to help Amazon's Inventory planing group to execute and analyze AB tests. We own a data pipeline that manages about 1.5 PB of data, with peak usage of 1500 EC2 hosts as EMR jobs. - Led the machine learning movement in IPCLab, pushing statisticians to learn CS aspects of ML and SDEs to learn the math behind ML models. Leveraging both off the shelf and homegrown models, reduced some deep dive procedure from several months to a few minutes. - Simplify and improve, I have helped reduced 90% of EC2 bill by relentlessly trimming fat. - Employee development, I have hosted 6-month long biweekly training session to train the statistician team to use Python and Elastic Map Reduce directly for their day to day job, significantly increased researchers' productivity. Amazon Cloud Drive: -Bootstrapped the Amazon Cloud Drive Native Clients team, Hired/Trained/Led the initial 6 person dev-team shipped Windows, Mac, and Web integrated clients for Amazon Cloud Drive in 6 month. -Designed and led the implementation of Cloud Drive Synchronization. -Helped Cloud Drive App on Kindle FireHD team to profile / optimize thumbnails cache performance. -Led numerous production stress tests, and invented various launching strategies to avoid / cope with sudden peak traffic. Ordering Pipeline: -Worked on the aggregation service behind Amazon.com checkout pipeline, unified checkout validation handling across about 40 teams. -Migrated about 3 million lines of legacy C++ code from RHEL3 to RHEL5. -Fleet capacity planning and Q4 readiness performance tuning

-Developed Virtual Storage Driver that simulates the "LUN Resync" feature in high end Fiber Channel SAN devices, allows feature team to test their features on desktop without purchasing many super expensive SAN devices. -Worked on the initial investigation of porting Exchange Server onto Windows Azure. Lots of low level hacking and mixed user-mode and kernel debugging. -The "Windbg guy" in the teams I worked and helped many other co-workers debugging all sorts of nasty bugs.