My Programming Projects

Pest Spread Prediction and Notification App - Anticipate

• Developed a mobile application that helps farmers identify and track pest risks across 6,731 agricultural fields in Yolo County.
• Integrated Gemini Vision AI and UC IPM vector search to identify possible pests from uploaded images and generate relevant reports.
• Implemented breadth-first search pest propagation and PostGIS scoring to estimate how pests could spread between nearby fields.
• Added proactive alerts across 63 irrigation districts using Expo push notifications.
• Supported offline uploads with automatic retries for users in areas with unreliable connectivity.

Cloud-Native Media Review Platform - Reel Feedback

• Developed a collaborative media review platform for uploading, watching, and providing feedback on video projects.
• Built the application with Next.js, Cloudflare Workers, and Cloudflare D1, with Google OAuth for secured access.
• Created an AWS Lambda and FFmpeg processing pipeline that converts uploaded videos into HLS streams.
• Generated four to seven adaptive bitrate versions between 1080p and 360p so playback quality can adjust to the viewer's connection.
• Reduced the time between uploading and playable video availability to approximately 7-20 seconds for short-form content.

IEEE ITAIC Published Research Paper

• Achieved 93.5% accuracy and 93.81% F1 on child vs. adult speech classification across two held-out cross-corpus test sets.
• Increased training data 4x to 31,936 utterances by unifying six datasets, including two private datasets from UCLA's Speech and Auditory Processing Lab.
• Implemented impulse response convolution augmentation to normalize recording artifacts and improve model robustness.
• Extracted embeddings using HuBERT, Whisper, and ECAPA-TDNN for robust feature representation.
• Benchmarked and optimized neural networks, logistic regression, XGBoost, and transformer models to identify the best-performing approaches.

FRC Robot Code

UCLA COSMOS Research Poster

• Enhanced a public dataset using impulse response convolution to simulate diverse recording environments and improve model generalization.
• Developed multiple AI models with TensorFlow to classify music genres in real-world contexts, achieving a 34% accuracy improvement over the raw dataset.

Recipe Scout

• Developed a privacy-focused web app that allows users to take a picture of ingredients and generate personalized recipes using in-browser YOLO for object detection, sending only detected ingredients to ChatGPT.
• Removed the need for external recipe APIs while producing high-quality recipe suggestions.
• Integrated secure Google OAuth authentication and stored user recipes in MongoDB.
• Enabled social features including recipe sharing, commenting, and feedback forums.
• Implemented privacy controls and shareable recipe links for public or private recipes.
• Delivered a full-stack AI-powered solution combining computer vision, LLMs, and interactive web technologies.

OctoScout

• Developed a modular scouting web application for FRC Team 9084, designed to adapt to any game, improving data collection, organization, and real-time match analysis.
• Created dynamic, interactive data visualizations using CanvasJS, enabling the team to identify trends, optimize strategy, and make evidence-based decisions during competitions.
• Established and led a scouting team in 2025 to maintain, expand, and iterate on the web app, ensuring it remains actively used and continually improved.
• Added features such as customizable reports, team comparisons, and performance trend tracking, making the app a critical tool for competitive success.

Webcut

• Developed a fully online video editor using Next.js, Tailwind CSS, and FFmpeg.
• Implemented features including cutting clips, setting in/out points, audio support, video resizing, and rotation.
• Enabled project saving, loading, and exporting at full quality, providing a complete end-to-end video editing workflow in the browser.

Slime Mould Simulation

• Developed a simplified simulation of slime mold behavior using HTML, CSS, and JavaScript.
• Optimized the simulation to efficiently run in the browser, handling up to around a million mold particles.
• Inspired by Blog post by Sage Jenson.

Issue Tracker

• Developed a customizable web-based project management program with an HTML, JS, and CSS frontend and Node.js (Express) backend.
• Implemented user authentication with Passport and data storage using MongoDB.
• Enabled features including custom projects, file uploads, and user management for collaborative project tracking.

OctoScout 2023

• Developed a customizable web-based scouting system for FRC Team 9084 for the 2023 "Charged Up" competition.
• Built a web user interface for inputting match data, which is uploaded to a server for collection.
• Stored data as CSV files to enable easy analysis and support team strategy decisions.

Electromagnetic Particle Simulation

• Developed a particle simulation as an AP Physics final project using HTML Canvas, JavaScript, and CSS.
• Implemented programmable particle charges and sizes, simulating electric and magnetic fields.
• Built the simulation from scratch, demonstrating fundamental physics principles interactively in the browser.

Platformer V2

• Developed a platformer game from scratch using HTML Canvas, JavaScript, and CSS.
• Implemented level loading and tracking of player times, with data uploaded to a server for competition among friends.
• Created an engaging game that was widely played and fostered friendly competition for fastest times.