* Scale up 10+ Millions QPS distributed system on critical path of content platform of Roblox

* Designed and build entire data ingestion pipeline from 0 of in Roblox with one of the largest matchmaking traffic in this world, including counters, logging, SNS/SQS data process, Elasticsearch, Hadoop/HDFS

* Design & build brand new ML online prediction services integration with legacy Net Core applications with ONNX. Horizontal scaled and optimized with low latency & high availability requirement

* Built experimentation platform, enable A/B testing and result monitor for scientists as easy as click buttons;

* Built core components of innovative AutoML time-series Prediction Platform in production environment, automates the entire time series prediction workflow in autoML way.

Technology including:

* Spring Boot, DASK, Vue, Express

* Redis, MySql, Spark Cluster

* Full-stack web application, Distributed Data pipeline, Microservices, AutoML

Tech Part:

* Designed and developed application Timbr, a large-scale Meteor application that turns an arbitrary web page into structure JSON data; plug-and-play APIs and interfaces for users without programming experience (i.e., automatic web crawler/scraper)

* Used Vue.js and D3.js on the front-end component of Timbr to visualize/collect user’s action toward target websites, provide view components and dynamically visualize the JSON data generated

* Used Node and Electron on the back-end component of Timbr to render pages and mimic user’s actions to crawl/sample web pages

* Deployed Timbr on both AWS and local servers to crawl/sample millions of web pages from sites such as eBay, Amazon, GitHub

Research Part:

* Focus on Markov Chain Monte Carlo(Random Walk) method for real world web data mining, sampling from hidden database of online social networks, graph theory and web information retrieval

* Make aggregate estimation over big picture of the datasets of online social network based on the samples we capture, e.g., Total number of the users in Facebook.

* Work on speeding up about 50 percent of convergence rate of the state-of-art random walk sampling method on both Graphical models and Bayesian Hierarchical Models

* Using revised Timbr showing the on-the-fly sampling and estimation process for random walk methods on real world websites, like Amazon and eBay

* Communicated with internal divisions and leveraged data analysis to optimize the average roadway operation time and decrease errors (e.g., missing data error, duplicate data error, etc. ) in MS SQL Server.

* Built pipeline (MS SQL Server and Python) to process real-time traffic event/accident data, and streamlined the workflow including data extraction, cleaning and exploratory analysis.

* Developed and automated dashboards and visualizations by using Tableau and Python packages (Matplotlib and Smtplib).

* Deployed Machine Learning models on traffic accidents data to better prioritize the CCTV cameras on monitoring screen for Transportation Management Center of DDOT.

* Created maps and layers containing hundreds of communication networks(routing , detection devices) using ArcGIS.