AI and Data Science Intern

Worked in the Consumer & Community Banking division to improve model interpretability by binning loan applicants and hierarchically clustering model features

Maintained an ETL pipeline for Consumer & Community Banking risk records and preprocessed 100 GB of raw consumer records for use across business banking loan models

Used Pyspark and XGBoost to improve the accuracy of our team’s small business loan default model by 12%

- My fourth internship

- Summer 2022

- Duration: 2.5 months

- Wilmington, DE

Summer Research Fellow

I was selected as part of a 16 person cohort to work on research in statistics and data science at Carnegie Mellon University.

Advised by Dr. Kostas Pelechrinis, Professor int he Information Science department at University of Pittsburgh, and author of 'Mathletics'. I contributed to a first of its kind analysis to predict expected goal count in soccer by using stochastic game dyanmics

Built a web application backend using Python integrated with a Javascript frontend to enable users to pit two soccer teams against one another and visualize the outcome of a match

- My third internship

- Summer 2021

- Duration: 2 months

- Pittsburgh, PA

Data Science Intern

Built an algorithm to identify when a Rolls-Royce AE3007 engine has reached its optimal compressor wash interval

Mapped 500K datapoints of Rolls-Royce engines to public commercial flights (across a 5-year span) to create a 1:1 engine to plane relationship when geospatially visualizing flight paths

Reduced the hours required to recognize which flights require compressor wash from 3-4 hours to less than 5 minutes by deploying a R Shiny application that structures, visualizes, and pinpoints hundreds of at-risk planes in real-time

- My second internship

- Fall 2019 - Spring 2020

- Duration: 10 months

- Remote

Information Security Intern

Visualized and investigated data transfer cybersecurity risk from company 3rd party vendors

Cleaned 87 MB of LPL user portal data with Python and transformed cataloged logs into multiple workable data frames to identify instances of malicious access on the company website

Presented a risk assessment report to senior executives highlighting frequent gaps in data privacy regulations

- My first internship

- Summer 2019

- Duration: 3 months

- La Jolla, CA

Make-A-Monet

I worked on creating an application to convert everyday images (such as nature scenes) into paintings

Using Pix2Pix and Cycle GAN architecture we were able to achieve image style transformations to generate "Monet-esque" paintings from user input.

To make our application user-friendly we hosted our work using Heroku and built our front-end with the Plotly-Dash framework.

Dog Breed Predictor

Using TensorFlow I created a web application with the goal of predicting a dog's breed using an image.

Capable of predicting 121 dog breeds, this web app is built with the help of Inception v3

On top of the breed clasifier, I also created a page within the web app dedicated to recommending dog breeds to a user based on owner-preferences (i.e. amount of maintenance, dog size, ability to be trained).

Variability In Baseball Stadiums

Among baseball enthusiasts there is a prevailing theory that the dimensions of a baseball stadium affect batting statistics.

Nowhere is this more true than with The Green Monster - the left field wall at Fenway Park that is famous for preventing home runs on many line drives that would clear the walls of other ballparks.In order to investigate trends like these, I use R Shiny to visualize how hit distance, effective speed, launch speed, and bases covered vary between all major baseball stadiums in the United States.

A Comprehensive Analysis of COVID-19 Vaccine Rollout

I led a team of three other students to investigate vaccine rollout across the nation. By investigating vaccine success state-by-state, we aimed to explain the disparity across the country.

I used Facebook's Prophet library to look at different milestones concerning vacccination success, including President Biden's 100-day plan.

On top of the analysis of vaccine rollout, I also used NLP to interpret the twitter sentiment surrounding different COVID-19 vaccines, international politicians, and other global entities.

Predicting 2021-2022 NBA Salaries

I worked on creating a salary predictor for NBA players. Using attributes like a player's position, age, turnover rate, and games played we sought to find out how much a player might make in their next season of play.

After using linear/polynomial regression, support vector machines, and random forest, I landed on using the gradient boosted framework, XGBoost.

I developed a web application, and built the final model by splitting data into Rookie/Veteran groups and running gradient algorithms on different subgroups.

An Investigation of the California Wildfire Crisis

As project lead, I led a group of five others to create a comprehensive article for DataRes at UCLA.

This article looks at California wildfire growth over the last decade, and the role climate change has played in exacerbating the current wildfire crisis.

I employed a variety of geospatial python libraries in order to best capture the existing patterns and characteristic attributes of California wildfires. All visualizations were created from scratch using CAL FIRE data and APIs.

Forecasting Hot Dog Sales

Every year, thousands of hot dogs go to waste across Texas due to miscalculation of demand. We sought to help Chevron prevent this by extracting insights into consumer behavior and creating a model that can accurately forecast sales.

We utilized machine learning to create a cook plan for certain buckets of time, saving Chevron money. We acheived a root mean squared error (RMSE) of 3.55 against new data (the lowest RMSE across 45 teams).

This project was completed as part of the 2021 Rice Datathon and my group won 1st place in our category.

Virtual Assistant for the Visually Impaired

In this project my team built a virtual assistant to convey COVID-19 information audibly and over SMS.

We noticed a gap in catering for those with visual impairments and addressed this issue with the tremendous capabilities of speech recognition.

We essentially created a virtual assistant that can perform a variety of tasks. These tasks range from answering questions about the virus using an AI-powered chatbot, to enabling quick-and-easy dissemination of real-time COVID-19 county-specific updates.

Retina AI Datathon

The aim of this datathon was to gather eCommerce insights and leverage analytics to diagnose business strengths in terms of customer behaviors.

I was charged with analyzing visualizations that projected customer lifetime value (LTV) over 5 year and 10 year periods.

By analyzing specific quantiles of our subset, I came to the conclusion that my client (the eCommerce company in question) should target higher spenders and incentivize to those customers who fall in the top 60% of high-spenders. I won 2nd place overall.