Welcome to My Page

Hello there! My name is Leon. I am currently a Master’s student studying Data Science at Duke University. I received my Bachelor’s degree in Chemical Engineering from the University of Washington.

During my undergraduate studies, I realized the importance of problem solving through data-driven approach. That’s why I have decided to combine my previous learnings with data science.

The past few years of research and project experience in machine learning, computer science, and engineering has taught me about solving difficult problems in interdisciplinary fields. My dream is to apply data science to places where help is most needed.

Recent Work Experience

May 2021 - Aug 2021

Data Science Intern

@ Windstream

Facilitated proactive live customer services through sentiment analysis by building, training, and packaging NLP transformer models for chatbot systems in PyTorch.

Established access to topic classification and sentiment analysis models across different platforms through model deployment by programming REST API endpoints.

Accelerated customer digital adoption by providing data-driven suggestions and building insight dashboards on user activities, remedy tickets, system message errors, and customer satisfaction using SQL, Python, RapidMiner, and Grafana.

Realized quality A/B testing through building census demographic lookup of the customer base from randomized grouping.

Jan 2021 - Present

ML Software Programmer

@ Duke Health

Assist doctors in finding effective treatment solutions through modifying existing Bi-Clustering algorithms in R and Python, which match patient demographics with cancer symptoms.

Researched and tested Bi-clustering algorithm applications in symptom patient study to help professor propose R01 grant.

Jan 2019 - Jun 2020

Research Assistant

@ University of Washington

Devised a deep learning model - variational autoencoder with the research team to explore chemical reaction pathways and predict intermediate chemical species using TensorFlow.

Built a training dataset by computing all possible chemical species from reaction pathways using Python libraries such as NumPy, Pandas, etc.

Implemented molecular rotational techniques with quaternion coordinate system in Python and C++ to visualize molecule movement in space and study interactions with different interfaces.

Projects

Feb 2021

Algorithmic Stock Trading

Algorithmic trading executes trade orders based on pre-programmed instructions without human intervention.

We built several stock trading agents utilizing both traditional indicators and machine learning algorithms. Some of the notable models that we built include LSTM, ARIMA, Deep Q Learning, and Evolution Strategy. Our machine learning methods fared better than our technical indicators in most cases.

Oct 2020

Duke 2020 Datathon Winner Submission

What are the long-term, global economic impacts of the Covid-19 pandemic? Our team recently tackled this pressing question at the 2020 Duke Datathon.

Our team leveraged population, economic and Covid-19 specific data to predict the impact on key economic indicators. Using a time series model, we predicted how the countries would perform in the forseeable future.

Aug 2020

Movie Recommendation Web Application

We built a clean and user friendly web interface that provides users with movie suggestions based on movies that they pick.

The website is built using Flask application with NLP algorithms supporting the backend computations. It also utilizes cloud services and setup continuous integration and deployment allowing us to update and make changes effectively.

Apr 2020

Modular Ammonia Synthesis Process Design

We designed a small scale ammonia synthesis process plant for farms. Ammonia is widely used as fertilizers in farm. However, manufacture of ammonia is an energy intensive process and is typically produced in large scale plants far away from farms.

Designing a small-scale ammonia synthesis plant in geographic proximity to the customers in the Midwest is ideal in reducing transportation costs, providing an alternate supply chain for these customers, and meeting the large ammonia fertilizer demand in the locals.

Mar 2020

Pneumonia X-Ray Detection

In this project, I built robust convolutional neural networks to detect patients with pneumonia disease with over 87% test accuracy from X-ray images using Tensor-flow.

The significance of this project is to help doctors quickly and accurately diagnose patients with pneumonia, which is a common symptom of COVID-19. To make it convenient for more people to use, I built a user-friendly interface to give more people access to this disease detection tool. With a simple click of the upload button, one could get the prediction result in just a few seconds.