Selected works
Projects built around real questions
I enjoy projects that enable me to learn a new domain, interpret messy data, and transform the result into something people can understand and use.
Deep learning
Generative AI
Research
Software
Machine learning · Climate research
High-resolution NO₂ mapping
Tracking global NO₂ levels matters for understanding air quality and climate change, but generating high-resolution measurements with WRF-CHEM is computationally intensive. With UC Berkeley’s Cohen Research Group, I helped lay the groundwork for a machine-learning alternative.
I aligned more than 47,000 WRF-CHEM pixels with NASA OMI pixels and engineered geolocation data to improve correlation, streamline analysis, and prepare the dataset for model training.
Game development · Procedural generation
Poke-Man
I designed and developed a multiplayer game that combines Pac-Man-style gameplay with the Pokémon universe. Players control Pokémon characters through procedurally generated mazes while completing objectives and avoiding obstacles.
I added persistent game saves with Git, intuitive menus, and a dynamic heads-up display while learning the practical challenges of real-time multiplayer mechanics and game UI design.
Created for educational purposes. Pokémon character designs belong to Nintendo, Game Freak, and Creatures.
Natural language processing · Deep learning
Part-of-speech tagging with BiLSTMs
Part-of-speech tagging identifies the grammatical role of each word in a sentence, giving language systems essential context. I built a tagger with a bidirectional LSTM trained using mini-batch stochastic gradient descent and pretrained GloVe embeddings.
I incorporated dropout to reduce overfitting and experimented with model configurations using the 500,000 most frequent 300-dimensional word embeddings.
Sociolinguistics · UC Berkeley
How people say “yes” and “no”
People in the same linguistic community can communicate the same idea in strikingly different ways. I studied variation in confirmation and declination responses among English speakers at UC Berkeley.
Over two weeks, I collected auditory linguistic data and analyzed how response variants related to gender, race, relationship formality, and situational context. The experience connected quantitative analysis with the social texture behind the data.
Full-stack web app · OpenAI Build Week
Danz World Cup League 2026
I turned a friend-group World Cup prediction pool into a responsive web app with Google sign-in, group and knockout picks, bonus questions, automatic scoring, and a public leaderboard.
The Firebase-backed app handled real players and tournament edge cases, including kickoff locks, hidden picks, extra-time scoring, admin grading, transparent point breakdowns, and a final winners podium.
Sports modeling · Interactive data product
US Open Forecast
An interactive men’s singles draw where fans can pick every remaining match, carry winners through the bracket, and compare their champion with a data-driven tournament forecast.
The experience combines the official 2026 draw and Central-time order of play with bracket-aware title probabilities, one-click projections, responsive match cards, Google sign-in, and device-local saves.
Want the one-page version?
See the experience behind the projects.
Open my resume
