Hack@Brown is Brown University’s largest annual hackathon, attracting hundreds of hackers from across the country every January. For the 2020 hackathon, my co-lead and I led the development of the hackathon’s website.

I interviewed and hired 5 additional members, set up the project structure and Git workflow for the team, and guided them through the construction of our landing pages, registration infrastructure, and admissions dashboards. This required me to maintain a birds-eye view of a complex project which involved React and TypeScript on the front end, Firebase for authentication and file management, and a Flask REST API on the back end. I was also responsible for all domain management for hackatbrown.org and automating deployments to our development and production environments.

Leading the Dev Team was a new experience for me, and I’m deeply grateful for the opportunity. Not only did I get to lead the development of a widely-used website, but I got to work closely with the talented people in our organization, such as our design and sponsorship teams. I also learned how to evaluate and communicate tradeoffs with members in the organization under tight deadlines.

Unsupervised Representation Learning in Atari is a paper published by Mila which proposes a new representation learning method for Atari games. Their method, ST-DIM, learns a representation of the game by maximizing the mutual information between subsequent game frames.

My team and I reproduced this paper as part of the NeurIPS 2019 Reproducibility Challenge, which we participated in under Professor Michael Littman’s class, “Learning and Sequential Decision Making.” In this project, I built and trained the base encoder architecture (a convolutional neural network), researched ST-DIM and its prerequisite papers, wrote and revised the report, and presented our findings. This was my first time conducting research, and I learned how to read (and re-read!) papers thoroughly and how to evaluate model performance in a scientific manner.

Partners: Lawrence Huang, Kendrick Tan, Shekar Ramaswamy

Elephluent is a language-learning platform which helps children pick up new languages through activities like matching games, speech-based exercises, picture learning.

I designed the site with Adobe XD, where I drew inspiration from apps like Duolingo and PBS KIDS. I also implemented the front-end with HTML, SASS, and jQuery. This required me to collaborate with the rest of my team to send lesson and user progress data via a Java-based REST API.

Partners: Dat-Thanh Nguyen, Kristen Mashikian, Shekar Ramaswamy

Manim Interactive Docs is a website I built to document common use cases for manim, the math animation library created by 3Blue1Brown. I created this while working with Professor Samuel Watson to help myself understand the basics of the library. I first wrote an animation for each “math object” (mobject) and rendered the video with Python. Then, I wrote the site to jump to the timestamp associated with each “mobject,” and displayed the Python code snippet that would render it in manim.

One of the coolest parts of this project was how lightweight it was. Using the Vue.js library, the whole site only took ~200 lines of code.