GoGreen: Carbon Tracker

  • Released iOS application to accurately measure the carbon footprint of a household given monthly energy consumption
  • Publicized GoGreen through social media and environmental organizations, driving 1000’s of downloads in 20 countries
  • Collected data and set up database infrastructure to show users how they compare to others in the country, state, and city
  • Engineered serverless architecture using Firebase and Cloud Functions to simplify app logic and increase app flexibility

March Madness with Machine Learning

  • Trained Support Vector Machine to build March Madness bracket, beating 95% of submissions in ESPN 2019 Challenge
  • Explored historical NCAA March Madness data via Jupyter Notebook to select strong features with which to train model
  • Benchmarked model against strategy-based brackets (random, high-seed wins) to determine the effectiveness of training

Viola-Jones Object Detection

  • Implemented Viola-Jones Object Detection algorithm in Python based on the description in the original research paper
  • Tested implementation on the CBCL face database, achieving 93% accuracy by tuning model hyper-parameters
  • Helped 3K+ people learn the algorithm by publishing two-part tutorial series in The Data Driven Investor


  • Designed algorithm to optimize user’s schedule given work preferences, availability, and coursework, resulting in manageable schedules for procrastinating students
  • Managed team through rapid 6-week development process by holding weekly debriefs and enforcing deadlines
  • Presented Kozo at the Mobile Developers of Berkeley Fall 2018 App Fair to hundreds of students, driving installs

Library Occupancy Prediction

  • Discovered sensor inaccuracy in UC Berkeley libraries after conducting exploratory analysis on time series occupancy data
  • Designed LSTM model to predict daily occupancy by training on derived measurements resistant to sensor inaccuracy
  • Built website with Flask to display predicted library occupancy, winning 2nd place at Innovate Berkeley Hackathon

Machine Learning from Scratch

  • Implemented Machine Learning Algorithms from scratch in Python with no machine learning libraries
  • Derived gradient-descent for Multi-Layered Perceptron, SVM, Logistic Regression Classifier, and Linear Regression

Fruit Classification

  • Designed deep convolutional neural network with Tensorflow to classify images of fruits with 92% accuracy on 40 fruits
  • Iterated upon the model design to increase the number of recognized fruits as well as accuracy

Cube Swarm

  • Collaborated with a friend to develop an addicting mobile arcade game for iOS and Android in Unity C#
  • Engineered the mechanics of the enemies and handled iOS-specific code