1️⃣Function as a principal data scientist and machine learning tech lead in the vacancy of an in-house principal data scientist, technically oversee marketing, call center, NLP, recommendation projects, and lead a global team of MLEs and data scientists in the US and China at its largest. 2️⃣Build company recommendation system for both call center and online from scratch:-For call center, in POC period, the prototype beat known 3rd party healthcare plan recommendation services on the market. In production in A/B test, it achieved 6% lift in 3 month plan retention rate compare to rule based plan selection. In production in full rollout, it reduces $70M cost on call center agent answering time during Q4 2020;-For online, in production in A/B test, it achieved 5+% conversion lift. When fully rolled out, it contributed to 58% revenue growth YoY in Q3 2021 compared to the same period in 2020. -The recommendation system as a key revenue driver was featured by the CEO on eHealth Q3 2021 earning call (https://apple.news/AGkOwnoB8TKST59TRTCh3dw)and eHealth 2021 Investor Relationship Deck (see attached) at, for example, RBC Capital Conference and H.I.G. Capital ($225 Million investment). 3️⃣Hands-on coding for deep learning/machine learning projects: -First in the world to apply Rectified Adam optimizer on HorovodRunner enabled spark clusters for distributed deep learning training; -BERT etc for other projects. 4️⃣Design and architect end-to-end deep learning/ machine learning system and pipeline: -Distributed training takes place in Spark with Hyperopt or HorovodRunner-Off line batch process in Spark or Snowflake and online cache in Redis-Low latency restful API call in sageMaker, with flask and gunicorn enabled multi threading, and asynchronous Kafka broadcasting. 6️⃣Represent eHealth with 3rd party:-Due diligence technical evaluation of corporate merger and acquisition, and vendors;-Solution architecture meetings with AWS and Databricks.

1️⃣Architected and designed the entire the machine learning algorithms for the backend recommendation system for all the use cases on Fanatics platform managed over 300 sites and million+ SKUs. -A unique challenge for sports merchandise is customers only buy the next level of sport equipment but rarely in reversed order. I built an RNN based model for this use case that out performed preceding collaborative filtering based supervised or non supervised models. -Built a CNN based model that was originally used for product image evaluation as part of recommendation system. Thanks to its effectiveness, it was then widely used by marketing and inventory departments. 2️⃣In 2017, I was perhaps the first one in the world to figure out how to make deep learning models trained on Keras to be served in a distributed fashion on Spark slave nodes, 2 years earlier than the open source MLflow pyfunc came out to achieve the same functionality. Note that MLflow came out in June 2018 initially for model tracking, and pyfunc spark_udf that supports Keras released in about 2019. My presentation at Tech Talk Speaker at Data Platform Conference in Apr 2018, 2 months earlier than the initial release of MLflow, demonstrated my way of distributed deep learning model prediction on Spark. 3️⃣In the pre-Mleap era, serialized and deserialized all kinds of models from R, ski-learn, spark ml writing my own codes for use in various languages in real or near real time. 4️⃣ Joined Fanatics when it was an extremely lean startup at Plug and Play in early 2016. Wrote the first draft of China business plan completely out of my daily job function and under direct supervision of the then CEO which contributed to Hilltop Capital’s $1B investment. My recommender contributes to the following: -Fanatics started with about 250 million valuation in 2011;-4.5 billion valuation September 2017;-6.2 billion valuation August 2020;-30+ b valuation 2023.

1️⃣Started as a UI design researcher but couldn't keep up with the work load of reporting UI a/b via the old procedure, even I worked 7 days a week. So I single handedly built the company's entire UI A/B test reporting pipeline and dashboard system (batch update) with HIVE, Redshift, Spark, Tableau, D3, Highchart, etc for the then world's 2nd largest e-commerce site with terabytes of daily log data in peak season.2️⃣Wrote a personalized expected shipping arrival date prediction model for onsite display. Worked with the business team and designed the entire visual display with logo and slogan of the "Quickship" product, that uses my model in the back end. 3️⃣Contributed to the overall personalized shopping project (pricing, NLP, review) with various technical and mathematical innovations, part of which are listed in the 3 patents/applications i have. Receive the prestigious company-wide inventor of the year award.