I worked as a teaching assistant of course Computer Network at CMU, Computer Science Department. During my work time, I updated the Content Distributed Network project with two new checkpoints about TCP load balancer and provided an implementation using consistent hashing and load-aware balancing with an JFI higher than 0.8.Moreover, I also developed video chunk load generator using Python requests library and load monitor using tcpdump tool, which supports the automation of testing using JFI as a metric among various load patterns and network conditions. Show less

I was responsible for a metadata management system, which was a web project for the purpose of managing metadata for hive database and I focused on web backend development including extraction of hive database using Java Database Connector, management of Mysql database using mybatis, and cluster-type management of timed tasks using Quartz. The whole backend project was based on Spring framework with SpringMVC Structure, and it has been launched by now. With the web frontend developers’ help, it could provide functions of displaying, editing and synchronizing metadata for the users.I also configured automatic periodical data synchronization on cluster and optimized synchronization time to one-tenth by optimizing logic and use multiple-thread. It takes less than 5 seconds to synchronize hundreds of tables in the final optimized version. Show less

I'm working at Pony.ai as a software engineer at map team, focusing on the offline localization evaluation platform. My work is actively involved with the whole data pipeline of localization data, ranging from preprocessing, ground truth generation, metric collection, data storage and also visualization at front end. By contiously refining current platform, I'm striving for a much compelling and reliable evaluation platform with aglity and precsion.

As part of my program requirement, I am working with my teammates on Project Felucca which is a cloud based execution platform of an existing command-line toolset called Pharos. Our goal is to provide an execution platform that could unify the Pharos toolset and make it easier to share and distribute. With Felucca, the users could upload their files and then execute the complex batch job of pharos tools using a user-friendly page and get all the stdout and output files easily.Our Felucca web application acts as a control center and execution platform of Pharos tool set, supporting batch task submission, real time output streaming, execution isolation and file durability. We also achieved toolset scalability and machine scalability by providing docker-based abstraction layers for toolset and execution nodes and used scheduler to orchestrate the executions. Show less