data analyst & software engineer

• Sign-In / Sign-up: A whole set of secure sign-in & sign-up with required HIPAA consent form
• Email Verification: Identity confirmation via secure SMTP link
• Medication Dashboard: Displays medications, logs, and statistics
• Medication Management: Add, delete, or update medication details
• Weekly Reports: Visualizes adherence with downloadable PDFs
• Help Center: Searchable FAQs with instructional videos
• Settings Page: Update profile and manage data-sharing preferences with caregivers

This portfolio website is my way of showcasing my skills while diving into new challenges and learning along the way! I discovered how powerful and flexible Tailwind CSS is, using it to create fun animations and unique styles effortlessly. I also focused on making the site accessible and responsive, with features like a navigation bar that magically adapts to any screen size. This project has been such an exciting journey, blending creativity and problem-solving to build something intuitive, inclusive, and truly me!

• Responsive design ensuring seamless functionality across mobile and desktop devices
• Image carousel for showcasing projects with smooth transitions
• Interactive hover animations enhancing user engagement with modules
• Adaptive hamburger menu and side navigation bar for compact screens
• Intuitive navigation with tags directing users to specific page sections

Driven by my passion for big data and a desire to simplify complex processes, my two-term internship at Sumitovant Biopharma was an incredible opportunity to transform how data analysis is done. I developed an in-house tool in Python and R (connected with rpy2) to streamline network meta-analysis (NMA) on clinical data, ranking treatments across multiple studies. The tool tackled challenges like inconsistent treatment names using predictive similarity scoring with 95% accuracy and automated data transformation. It also integrates proper analysis model for different data types (binary, proportional, numerical, etc.) and delivers results within seconds, making data analysts’ work faster and more efficient.

• Extract and process data from hundreds of ClinicalTrials.gov XML files
• Predict and standardize treatment names to align with the Cortellis drug list
• Automate the generation of comprehensive modeling result in one click

This news search platform, built with React and Next.js, brings the latest New York Times articles right to fingertips through the NY Times API! Beyond just browsing, it lets users save their favorite articles and offers a seamless, personalized experience. When people are logged in, the platform—powered by Firebase for web analysis and Firestore for the non-relational database —remembers the search history, which users can edit anytime. It’s designed to make revisiting topics or rediscovering past interests effortless and enjoyable.

• Search functionality to retrieve New York Times articles by keywords
• Dedicated user portal for saving and managing favorite articles
• Provide redirection options to access the original news source

This is an Instagram clone project for the EECS485 course in Umich, the first course for me to learn how to build a web application. Creating this was an amazing learning experience to develop both my front-end and back-end skills. React was used to allow for client-side dynamic pages on the home screen of Instagram and also for individual posts. Python and Flask were used to construct REST API, Flask cookies were used to store usernames and to authenticate users, and SQL was used for database access and storage. The web app was finally deployed with an AWS EC2 instance.

• Creating accounts, following, liking, commenting, uploading photos, and more
• Double-click on an unliked image to like it
• Immediately show new likes and comments
• Infinite scroll by displaying the next 10 posts when scrolling to the bottom of the page

As part of the UMich Multidisciplinary Design Program sponsored by ProQuest, we developed a machine learning model to predict the likelihood of books winning awards, designed for deployment on Rialto, an academic marketplace for data-driven librarian decisions. Tackling challenges like 0.03% award-winning books in a 10M+ dataset and 80% missing data for numerous fields, we invested significant effort in merging datasets, handling missing values, and transforming raw data into usable insights. We analyzed 6 data sources and there were a total of 75 features, among which we selected or used to create 6 features for our final model. Through rigorous preprocessing, time-series analysis, and Random Forest Regression, we achieved a 31% match rate between predicted and actual top 100 books, highlighting the strength of our data preparation and modeling pipeline.