Boston, MA

August 2022 - September 2023

Working full-time at Verisk, I had the opportunity to work on various tangential projects. These projects ranged from using a Random Forest Classifier to determine important features for building durability to conducting model validation using claims data to implementing building codes into our model. I also developed an efficient algorithm to quantify the nearest buildings, vegetation, and roads using satellite imagery. Using a K-D Tree, this algorithm could quickly compute distances and could be applied at the city, county, state, and US mainland scales. These efforts were pivotal in taking our US Wildfire Model from alpha to beta to final release.

Remote

June 2021 - August 2021

During my internship with Verisk, I worked in the Cybersecurity Research Team and developed a heuristic, rules-based model to classify whether companies were real data providers (ie. AWS, Azure) or not. In addition, I worked with the centralized Verisk Data Science Program to develop a classifier using NLP and Yelp reviews to determine if businesses have licenses (achieved ~96% accuracy). Finally, I placed 2nd in a company-wide hackathon by utilizing a Latent Dirichlet Allocation (LDA) topic model to uncover "hidden" topics; I did sentiment analysis on these topics to help predict insurance claims losses.

San Diego, CA

January 2020 - October 2020

At Intel, I analyzed global CPU usage and performance experiences using daily processor data from Intel (~250GB dataset). With simple explanatory data analysis, I discovered that desktops spend 20-40% more time in turbo than notebooks, especially at higher core concurrency. I then proceeded to do a root cause analysis using a Decision Tree Classifier and explained 50% of the turbo behavior variance with a handful of features. Finally, I presented my findings to upper Intel management and helped PC designers better provision systems to enhance the user experience.