Working under Professor Yoonjin Won and Youngjoon Suh:

• Computer vision monitoring micro-scale heat transfer phase change

Working under Professor Sandi Irani and Shion Fukuzawa at UCI:

• Leading a team of 6 creating a web-based simulation of the CHSH Game as an educational tool to teach CS 166 - Quantum Computation and Information.

Working under Professor Thomas Yeh and Professor Alex Ihler:

• Researching ML Acceleration algorithms and exploring combining AFP, a novel floating point quantization, Lottery Ticket Hypothesis Iterative Pruning, and ZFP compression for efficient matrix computations.

• Web Lead: developing iGPT--a ChatGPT-like web app used in our experiments with ICS 31-32 (Intro to Programming in Python) students to study the impacts of LLMs on CS education. Used in classroom sizes of 150+ students.

• Github 🔗

Open-source collaborative of over 70 student developers creating several cross-platform apps to enhance the lives of UCI students--with projects ranging from course schedule planning, historical grade distribution tools, and dining hall menu trackers.

Across all apps, supporting up to 20K+ total users, with 5K to 15K+ active users per week.

(Github 🔗)

Lead for the ZotnFound Team 🔗: An interactive lost-and-found social network app designed exclusively for UCI students and faculty on campus, with over 200 active users.

Organized the 3 largest hackathons in Orange County--ZotHacks, IrvineHacks, and VenusHacks--with over 500+ attendees and $35k in funds.

• Tech: Led a team of 8 building our club's home page; custom-themed sites for our 3 annual hackathons; and application tracking platform, supporting our intake of over 800+ applications.

• Marketing: advertising our hackathons and club events through social media with over 2000+ followers

• Misc: helped run weekly workshops and events with 30+ attendees

Arduino, Ngrok, React Native/Expo, Supabase, Raspberry Pi

An autonomous rover replacement of human "pacers" during marathons and long distance competitions

The Pacer Buddy app tracks/graphs the runner's pace in real time, records their runs' geolocation data on Google Maps, and allows a user to capture and upload photos of their runs

Demo Vid 🔗

Repos: Server 🔗 and App 🔗

Deepnote, Matplotlib, PyTorch

Studied the effects of applying ZFP compression to Adaptive Floating Point (AFP), a novel quantization method, for efficient ML acceleration, with varying levels of lossy compression, sparsity from Lottery Ticket Hypothesis (LTH) pruning, and input dimensionality.

Able to compress and quantize from FP32 to an average of 3.5 bits per element with 90% LTH sparsity, while maintaining a < 1% margin of error from the original baseline accuracy and reducing the # of bit operations by 98.81%.

Presentation 🔗

Paper 🔗

Neo4j AuraDB, FastAPI, Supabase Realtime/Storage, Next.js

A social network platform for campus clubs/students to post/view events using real-time pins on an interactive map.

Demo Vid 🔗

Devpost 🔗

AWS, Firebase, Google Maps API / Leaflet, React Native, SST/Ion, Chakra UI

An open source interactive lost-and-found network app designed exclusively for UCI students and faculty on campus, with over 200 active users

Check it out here! → Live Site 🔗

Google Colab, Pandas, Plotly, Keras

Compared the results of transfer learning on pre-trained models such as VGG-16/19, MobileNetV2, ViTb-16 Transformer, ResNet101 against classical CNNs on the COCO, CIFAR-10, and Kaggle datasets

Achieved a validation loss of 0.0058 and 82% validation accuracy on drone detection bounding boxes with ViT—a 40% improvement compared against other models

Presentation 🔗

Paper 🔗

Google Colab, Pandas, Plotly, PyTorch

Finetuned ViT and SWIN transformers ensemble for Chest X-ray (CXR) diagnosis with little to no radiologist intervention, achieving 87.2% validation accuracy on COVID dataset

Created Deep Convolutional GANs model to generate labeled CXR scans to feed into our transformers as training data

Presentation 🔗

Paper 🔗

C++

Created a custom heuristic-based Minesweeper AI solver using a frontier-based breadth-first search with probability calculation and pattern checking, based on surrounding tiles, to determine which tile to uncover next

C (programming language)

Spawns child processes for each shell command. Supports all Linux commands, background jobs, pipe and redirection operators, and custom shell commands

C++

Created an Othello AI Bot using adversarial game trees and alpha-beta pruning for efficient state space searching to determine the next best move against an enemy player

• placed 3rd in a tournament against other Othello AIs with a 95.2% win rate out of 284 games
• record (win/loss/tie): 270-13-1

Python

Physics simulator of several types of "simultons" created using Model-View-Controller (MVC) architecture

Juypter Notebook, Plotly, Pandas

Data exploration and analysis of the Montreal Crime Dataset from Kaggle. Presented data visualizations and findings to a panel of judges.

ExpressJS, PixiJS, Socket.io, React

Worked in a team of 4 to create Rattle: a multiplayer web-based rhythm video game with custom animated sprites and UI. Winning 2nd place out of 15 teams.

Pilot tested AWS UCI Credits portal + held an Amplify demo deploying full stack apps at IrvineHacks 2025

Organized Amazon Web Service (AWS) Certification Workshop for women in tech at UCI in collaboration with Women in ICS and UCI ELP.

Spoke on AWS cloud computing professional opportunities alongside guest speaker: Trinity Key, AWS Solutions Architect.

Slideshow 🔗

Organized Amazon Web Service (AWS) SageMaker Workshop for UCI students in collaboration with ICS Student Council and UCI Experiential Learning Program.

Spoke on AWS cloud computing professional opportunities alongside guest speaker: Mike Glover, AWS Sr. Solutions Architect.

Slideshow 🔗

Vercel OpenAI SDK, Next.js, Redis, ShadCN, NextAuth

• Ran a study on over 70 novice Computer Science students at UCI evaluating the effects of interactive prompting between LLM chatbots and CS students for better debugging, edge case testing, and programming education

• Developed iGPT, our custom interactive chatbot, in-house used during study

• Realtime Redis Dashboard to monitor chats during the study

• Published "Bridging Novice Programmers and LLMs with Interactivity" paper with our experiment's results to ACM SIGCSE Research Conference

Demo Vid 🔗

Created and delivered the ZotnFound's team End-Of-Year presentation, outlining our team's accomplishments, progress, and new production features.

Presentation 🔗

Architecture Diagram 🔗

Demo Vid 🔗