Projects

Projects

A mix of research, hackathon wins, and things I built to learn. A thread runs through the work I'm proudest of: engineering for accessibility, technology that helps people who are too often designed around.

Recognition: Outstanding Exhibit, UIUC Undergraduate Research Symposium 2025; Honorable Mention, URS 2024

ScribeAR: real-time transcription for accessibility

ScribeAR is a real-time transcription and augmented-reality captioning research project at UIUC, built to make spaces like classrooms accessible through live captions. I implemented a Whisper.cpp speech-to-text backend that brought transcription down to ~2 seconds per sentence, built a React hook for Whisper API calls (with controls for softness, speed, and history), and researched speaker identification, diarization, and audio classification.

Stack: OpenAI Whisper (Whisper.cpp), React, speech-to-text, diarization

UrbanHawx: facial recognition, live (HackIllinois 2022, Top 5)

Facial detection and recognition software that tells a safe, registered user apart from an unknown one in a live feed, using eye, face, and smile detection to differentiate people. Built in Python with OpenCV. Placed Top 5 at HackIllinois 2022.

Stack: Python, OpenCV, computer vision

Indoor Navigation for Visually Impaired Users

Indoors, there's no GPS to fall back on, so someone who can't see the room has very little to navigate by. This system reads the space through the camera instead: it runs real-time object detection to recognise doors, stairs, furniture, and obstacles, then speaks what's ahead and where it is so the user can move through an unfamiliar building on their own. I paired Mask R-CNN for detailed instance segmentation with a lighter SSD MobileNet model for fast on-device inference, and turned the detections into short, spoken cues rather than a screen full of boxes. The whole thing is built around one question: what does a person actually need to hear to take the next step safely?

Stack: Python, Mask R-CNN, SSD MobileNet, object detection, text-to-speech voice output