Work
Assistive Vision AI
Built an assistive computer vision system combining a custom-trained detection model, YOLO-based detection for additional street categories, OCR, and real-time audio guidance. The system identifies street signs, crosswalks, traffic lights, stairs, and hazards, reads text aloud, and was trained using a ~1,500-image dataset I built and annotated.
LazyQ
Developed and deployed the frontend of a virtual queue-management platform in HTML, CSS and JavaScript. Users scan a QR code to join a queue, track their live position and get a notification when it's time to return, with no registration or app download. Working with the team, I integrated the interface into the full web application, prioritising accessibility, responsive design and a frictionless experience.
Neural Representation Learning
A BCI pipeline for neural representations that transfer across recording modalities. It combines contrastive learning with EEG preprocessing, artifact rejection, provenance logging and channel-importance gradients.
View on GitHubRNN Music Generator
An LSTM trained on symbolic music to generate short sequences with learned structure and phrasing.
View on GitHubMachine Learning from Scratch
Core machine-learning algorithms implemented by hand to understand the mechanics beneath library abstractions.
View on GitHub