Zilong Yang

Wayfair, Release Engineering
Boston, MA

Software Engineer
June 2022 -- March 2023

• Worked in the Deploy Platforms team of 7 release engineers to maintain and improve Wayfair's engineering infrastructure tools, mainly Github Enterprise and Buildkite, to ensure the development and deployment processes within the organization are reliable, secure, and efficient.
• Introduced housekeeping workflows for Buildkite pipelines that run once a month to clean up abandoned or unnecessary pipelines and scheduled runs, cutting 1000+ unneeded computational hours and resulting in remarkable cost-savings.
• Built an extensible caching service for internal OpenJDK docker images and Maven dependencies with dedicated Buildkite queue and agents, accelerating the average pipeline runtime by more than 50% for relevant teams.
• Improved our Buildkite agent deployment process by automating the execution of Jenkins pipelines for image packaging and deploying, image functionality testing, error monitoring, and image rollback. This eliminates the need for human supervision during deployment, cutting ~50% of the time needed for this frequent run-the-business task.
• Handled and resolved multiple PagerDuty incidents during on-call hours that are related to services and tools upkept by our team, strengthening customer experience and software reliability by effectively collaborating with our customers during stressful moments.

Smith & Nephew, Sports Medicine
Andover, MA

Advanced Technology Development Co-op
September 2021 -- January 2022

• Independently designed and developed a deep learning U-Net model in Python for bone segmentation on computed tomography (CT) scans for the team's pre-market orthopedic surgery software
• The new solution to bone segmentation improved the accuracy of CT bone segmentation by over 50% and is able to handle a much wider range of edge cases
• Maneuvered medical images using 3DSlicer and Synopsys Simpleware ScanIP along with Python scripts to semi-automatically acquire corrected images and segmentations for training the deep learning model
• Identified the compared downfalls and advantages of using tradition image processing techniques versus using machine learning for segmentation on CT scans to justify the need for deep learning to further enhance the product's performance

Wentworth Institute of Technology, Digital Health Lab
Boston, MA

Machine Learning Research Co-op
January 2021 -- April 2021

• Collaborated with teammates to devise a deep learning network for an ongoing computer-assisted hand osteoarthritis (OA) detection software that automates the classification of OA severity using the Kellgren-Lawrence (KL) system
• Worked individually on researching and developing a deep learning model for hand segmentation on X-ray images as a vital preprocessing step which uplifted the overall accuracy of the hand OA classification pipeline from 75% to 85% by utilizing various Python frameworks to preprocess/postprocess X-ray images and train/evaluate the machine learning model
• Composed a medical abstract regarding the research conclusions, which was submitted to and accepted by the American College of Rheumatology Convergence 2021, titled "Automatic Hand Segmentation from Hand X-rays Using Minimized Training Samples and Machine Learning Models"

PROJECTS

360 Seating
Summer 2021

• Designed a full-stack web application that provides a complete online movie ticket ordering system with the additional "point of view" feature that allows the user to interact with a 360o view of their seats, aiming to ameliorate the stress on local movie theaters during the ongoing pandemic by possibly boosting sales of commonly undesirable seats
• Built extensively upon React for frontend, Express and MongoDB for backend, and other middleware and utility tools such as Cors, Morgan, Dotenv, and Mongoose to take steps forward in development speed, security, and robustness

CardChain
Fall 2020

• Deployed a decentralized, web-based training-and-fighting card game on the Ropsten Testnet using AWS Amplify to emphasize the team's understanding and the desirability of the slow transformation to decentralization with blockchain
• Applied smart contract concepts of the Ethereum blockchain using Solidity to develop an ultramodern procedure to imitate the database component in tradition model-view-controller design pattern without the drawbacks of a centralized data storage
• Incorporated state-of-the-art web development modules including React for frontend, web3 as middle- and backend, and truffle for local backend testing