Straightline
2026ADA compliance analysis platform — DiamondHacks 2026 winner
Built in 36 hours at ACM DiamondHacks 2026 (400+ participants). An ADA compliance platform that uses autonomous Browser Use agents to scrape accessibility data for real-world locations, Gaussian Splatting for 3D environment reconstruction, and Supabase for caching. Won 1st place for Best Use of Browser Use.
- –Won 1st place Best Use of Browser Use at ACM DiamondHacks 2026 among 400+ participants.
- –Built autonomous multi-agent pipeline to scrape and analyze ADA compliance data.
- –Integrated Gaussian Splatting for 3D reconstruction of physical locations.
- –Designed Supabase caching layer to persist scraped compliance metadata across sessions.
Browser Use · Gaussian Splatting · Supabase · Google Maps API · Python · Next.js
Fruit Recognition CNN
2025Real-time 24-layer GoogLeNet CNN — ACM AI Projects
Engineered a real-time fruit recognition system for the ACM @ UCSD AI Projects Team. Designed a 24-layer GoogLeNet-inspired CNN from scratch with custom IOU and loss functions, trained on the COCO 2017 dataset.
- –Designed 24-layer GoogLeNet-inspired architecture with inception modules.
- –Implemented custom IOU metric and composite loss function from scratch.
- –Trained and evaluated on the COCO 2017 object detection dataset.
Python · PyTorch · GoogLeNet · COCO 2017 · OpenCV · NumPy
Triton Robotics Autonomy
2025–20266-DoF pose estimation with solvePnP — Triton Robotics
Improved 3D object orientation accuracy on a robotics autonomy team by transitioning detection from bounding box methods to feature-based computer vision using sticker and sentry symbol recognition. Prototyped a Python/OpenCV node with contour detection and applied solvePnP with real-world object dimensions to achieve 6-DoF pose estimation.
Python · C++ · OpenCV · ROS2 · Computer Vision · Pose Estimation
Movie Recommendation System
2026Matrix factorization recommender — CSE 25, UCSD
Built a collaborative filtering recommendation system on the MovieLens 20M dataset using regularized Matrix Factorization trained with SGD. Designed a hybrid scoring model combining personalized latent factor predictions with item popularity signals, achieving Precision@10 of 0.0901 and outperforming both pure MF and popularity-only baselines. Explored the tradeoff between rating prediction accuracy and ranking quality in high-sparsity settings (98.6% matrix sparsity).
Python · NumPy · Collaborative Filtering · Matrix Factorization · SGD
Tigers Vote
2024SwiftUI voter registration app for ~400 students
Independently built and deployed a SwiftUI iOS app that simplifies the voter registration process for students. Shipped to ~400 students with a clean, native mobile interface.
Swift · SwiftUI · Xcode
RechargeTeach
2024Swift app for teacher burnout and wellness
A Swift app designed to help combat teacher burnout, featuring wellness check-ins, resource recommendations, and daily reflection prompts for educators.
Swift · SwiftUI · Xcode
PetHub
2025Next.js interactive anatomy learning app — WiC
An educational Next.js web app built with the Women in Computing project team. Features interactive pet anatomy explorations and learning modules, deployed on Vercel.
Next.js · React · Tailwind CSS · Vercel
Spotify Recommender
2026Collaborative filtering recommender — ACM AI Mentor
Co-mentored a team of 4 to build a Spotify recommender system using collaborative filtering and user similarity models. Evaluation methodology included Precision@K, Recall@K, and random/popularity baselines.
Python · Collaborative Filtering · scikit-learn · Pandas · Jupyter