ML Software Engineer

Handle data up to 5 billion events and 1TB a day; 30K-83Krps at peak; <10ms latency Coursera
- Machine Learning Certificate
- Udacity Deep Learning and Natural Language Processing Nanodegrees
Java, Python, Go, JavaScript, Erlang, Android, MySQL, PostgreSQL, Redis, Kafka, Riak,
Storm, Spark, Machine learning (CNN, RNN, GAN, RL, NLP…),
AWS (MySQL, Aurora, Redshift, Elasticsearch, EC2, S3, SageMaker, Glue, Kinesis, DynamoDB, Lambda, auto scaling group, ELB, Docker, ECS…),
Google Cloud Platform (Computer engine, Cloud Storage, BigQuery, BigTable, Stackdriver…)

GoGuardian

May 2017 – Present

Sr. Software Data Engineer

Working on data platform in a micro service environment. 1B events and 1TB data daily. 30Krps at peaks; machine learning.
Led back to school readiness campaign, for 5M users, with $3M annual spending.
Built a data deletion program.
Created an in-house data solution to ingest, dedup, aggregate, and persistent data, and also serves as a query engine for both time series and aggregation requests.
Optimized systems, improved performance, and saved cost; e.g., saved $20K/m on Aurora by removing gap locks, saved $5K/month on Docker deployment by tuning up JVM parameters.
Cleaned up the core database, removed 45.5% tables, 53.4% of data rows, 67.9% data size, in 3.5 months. Removed 4.18TB relational data and indexes. 5,400 lines of code changed.

Nordstrom Rack/HauteLook

Feb 2017 – May 2017

Sr. Software Engineer

Led personalization project. 44M users, 600K skus, 3 years of data…
Published a paper about A/B testing which proved prior experiments made wrong conclusions.
Identified correct personas; the existing personas were inaccurate.
Categorized products and customers.
Made business suggestions, say, blocked abusers.
Made a one-year plan to implement personalization on legacy systems.

Google

March 2014 – February 2017

Lead Software Engineer

Worked on AdWords and Android, the app is rated 4.3/5.0 in Google Play store and has 30,000 active daily users.
1.5M AdWords users are registered on it.
Monitored user activities and server metrics daily.

Cicadida.com

January 2014 – Present

Founder

GradientX

Nov 2012 – March 2014

Sr. Data Engineer

The company was bought by Amobee 18 months after its birth.
Designed, implemented, and owned targeting server and retargeting user tracking databases.
Developed a targeting engine that handled 5B+ transactions a day, 83Krps at peak time, with average 200 micron seconds processing time on each request (5K targeting/s), autonomously balance its workload, and never crashed in 6 months after its birth.
Tracked 480M monthly users using a Riak cluster of 25 servers and 9M hourly active users using 3 Redis servers on AWS; data access timeout is 20ms and the average is <10ms.
Data were compressed to 1/3 of original size so we saved 2/3 cost.
Redis instances on one EC2 m1.xlarge handled 20Kqps for hourly user activities.
Dumped and analyzed user data nightly and presented reports every morning.
Monitored production environment by creating a set of logging utilities to get real-time statistics.
There was no cost for monitoring.

TheBrain Technologies

September 2008 – September 2012

Software/Systems Engineer

Implement and complete new system design for market leading mind mapping tool; improve code quality; improve product reliability, performance, and usability; serve as both front and defense line of eliminating intricate software bugs
Optimized CPU usage, threading, and database operations, reduced 30% CPU time on Mac OSand reduced 1/3 threads in run time
Added keyboard accelerators into the messaging system.
User actions became much faster.
Integrated Windows and Mac OS native resources to Java program: provided search ability on Calendar/Address/Message on Mac OS, worked with Windows registry; and made screen savers using Java.

PhD. in Computer Science in Wayne State University
- Dissertation: Multi-objective Optimization Using the Cultural Algorithms
BE in Computer Science from Tsinghua University, Beijing, China
Publications Upon Request