As someone interested in data analysis and healthcare, I wanted to gain more experience with real-world data. I began analyzing Epic Systems EHI Tables exported from patient portals of myself and my family members (with their permission!). Detailed analysis is available here and I will add visualizations here as I develop them.

First, I used claim remittance information from the extract to estimate the 10th, 50th, and 90th percentile prices for each billing code in line with the 2026 CMS Hospital Price Transparency rule. I then used the remittance data and the Healthcare Utilization Project's Clinical Classification Software to determine what types of services were provided. I then created reports in Power BI.

Categories of claims:

Some sliceable tables showing price distributions by insurer and by billing code, with identifying information redacted:

While living in the DC metro, I frequently rode Metrorail, Metrobus, and other regional buses. Since transit around the DMV is operated by different entities, I often had to switch between multiple websites while commuting. Additionally, while WMATA's busETA has a map on desktop, on mobile it only shows a list of stops. ART RealTime similarly does not have a map and only shows distance to a specified stop. In March 2024, I made an R Shiny Markdown app that shows the location of multiple trains and buses I rode frequently using protocol buffers from the WMATA/ART APIs. (Note that due to the amount of data, the website may time out while loading.) The website has maps, calculates great-circle distance between buses and specified stops, and shows current WMATA alerts. I initially hard-coded specific routes and stops, then updated the app to allow the user to select a route. Updating the app also accounted for WMATA's updated bus networks in summer 2025. Mousing over each vehicle shows vehicle number, whether it's in motion, nearest stop, and crowding.

Legacy map with defined stops and estimated distance:

New maps with changeable routes:

In my first year at Duke, I began scraping transaction data from my student account to analyze patterns in how I spent my food points (Duke's dining dollars). By senior spring, I had enough data to write a narrative for the Duke Chronicle on how my spending habits changed over the years. I created Flourish visuals to accompany the story. One of the interactive visualizations is featured below.

This project received an honorable mention in the Multimedia Feature Story of the Year category in the 2023 Associated Collegiate Press Pacemaker Awards.

I contributed to the Duke Chronicle's annual first-year survey project for three years, leading the team in the third year. Each year, the team designed a survey in Qualtrics, disseminated the survey via flyers, email, and word-of-mouth, and presented the results through a series of articles. We received around 300 responses each year and compared our findings to publicly available information about the first-year class. Trends were visualized in Tableau.

Some of my reports:
Meet Duke's Class of 2026: The Chronicle's first-year survey results
Chronicle to release survey data on Class of 2025: Inside the methodology and limitations
COVID-19 pandemic impacted first-years' living situations, but many still found community

Independently, I designed and administered a survey on students' experiences getting the COVID-19 vaccine during the pandemic. The sample size was nearly 100 people and yielded some interesting trends.

Over the course of the fall 2021 semester, I developed a website to display North Carolina health statistics at the county level for my Database Systems class. In addition to generating visuals, the website could also calculate the nearest health facility available in the database. After the course ended, I updated the website to make it more user-friendly, with the option to create an account, save search results and make changes via an admin dashboard.

I wrote a Python script to analyze my text messages, which were stored in a db file on my Macbook. The script calculates how many messages I exchanged with a particular contact, which emojis and reactions I used most frequently, and how often I "double texted" before getting a response and vice versa. The analysis shown in the below images includes messages between December 24, 2020 and December 30, 2021.