Dance in ads decisioning / machine learning team of Disney+, Hulu, and ESPN+

Lead two teams focusing on ad decisioning

Lead a team of 4 engineers focusing on ad decisioning:* Rearchitect monolithic legacy ad system, which is the core of the ad decisioning system in Hulu, to a distributed microservices* Develop instrumentation system to collect business/key metrics, to maintain the health of the entire backend infrastructure, and to keep the uptime of the system above 99.95%

Lead a project to revamp VAST pipeline, which is the most resource consuming area in the ad server here:* move and re-architect resources heavy VAST process (i.e. pre-process, ingestion, metrics, etc.) from legacy services into micro-services to make it more sustainable and robust* design a data heavy monitoring system to govern the health and the life of each ad from its creation to the end of campaign* design UI for partners/customers to watch the health of their campaigns, for solution engineers to debug any discrepancy, and for product managers / data scientists to easily analyze potential improvements of our systemsIntegrate third-party ad systems with Hulu: * Work closely with partners (e.g. Disney, Fox, NBC, etc) and ad server teams (e.g. Google Ad Manager, Magnite, etc) * Gradually tear down the current legacy systems into more adaptive and maintainable micro-services in order to integrate with more third-party ad or measurement systems.

Worked on ad serving system, with thousands of advertisers and near a billion ads served per day. Some of the major focus areas were trafficking and quality control (e.g. ad pacing, ad targeting, viewer experience), to deliver maximum value to the advertisers while maintaining and engaging overall experience for Hulu viewers.Started a proof-of-concept on migrating legacy services into Cloud, which involves: * experiment on using Terraform to quickly and easily deploy applications to Cloud * building a micro-service CICD deployment pipelineThese steps were all to prepare for slowly digesting the huge legacy service into more maintainable codes and newer technologies. Eventually, the project went successful and still in-progress.

Led a team of 4 engineers to build a scalable time-series system/storage for monitoring metrics that can handle the load from tens of thousands of servers, able to handle over millions of data points per second. Successfully transitioned a proof-of-concept (of technologies chose to be used) project into production state with limited capital resources. The success of the project brought more attention (i.e. people and resources) interested and involved.As the owner of the project, the roles are: * Research: design the system from scratch and prove the selected technologies are capable to accomplish the needs * Backend: develop the system in Go, tune the databases (Redis and Cassandra/DynamoDB) as needed, and deploy/manage the clusters * Frontend: design and develop the client-facing API as well as the user interface (based on an open source library) for visualizing the received data * Scrum master: leading and mentoring onboard engineers to work on the project

* Enhance the monitoring system for all RightScale managed servers and develop its user interface in Ruby* Develop a complex billing system to track customers' usage on the cloud services in Ruby* Decouple frontend templates from backend data sources using Backbone.js