Working on some magic stuff 🤖 - Custom Q

Pull Through Cache Service for ECR launch in 2021 re:Invent• Worked on the project let customers have their own cached copies of the images from an upstream public registry in ECR. • Excited to be the speaker to introduce Pull Through Cache service in re:Invent and help customers to onboard it onsite.• Maintained and develop new features post PTC launch.Zonal Resiliency for ECR • Focus on the resiliency of ECR as part of Zonal Isolation goal, to provide better availability in the face of a zonal outage or grey failure.• Implemented a set of infrastructures and APIs connected with ECR core flow, build up a service to diagnose & weight away from an impaired zone to keep service available.QLDB to DDB Migration for ECR Images• Regarding the increasing demand usage for ECR, migrate all resources from JournalDB to DynamoDB to provide high service performance for the customers in core push & pull workflow.• Responsible for the design, implement and roll out of Image Count, Recovery and Reaper parts migration work.

Design, implement and deliver the Route53 host vip mapping in B16• Support BWDaasMappuller grab mapping data from S3, update map and reduce fromexisting 5 second pull to 2.5 seconds.• Replaced functionality of Snooper to convert mapping data to appropriate mitigation config byintegrating into BWDaasMappuller.

Designed the structure and logic of the Dynamic PO index project, supported it in two versions database.Wrote a design doc, implemented PO index functions with backfill job, to Spanner when clients update new index config using Flume and Mutation Pool, optimized CPU consumption by about 50%.

Import yearly data with ETL, implemented error-rebalance algorithms and Neg-Pos Failure algorithm. Utilizing various Machine Learning methods Random Forest, SVM to train and test models and export predication curve lines.