Final year project detecting driver yawns from video. MediaPipe mouth landmarks give a per-frame aspect ratio, hysteresis thresholds segment that signal into discrete mouth-opening events, and six temporal features per event feed a Random Forest classifier. A real-time webcam mode reuses the offline feature code directly, so live inputs cannot drift from what the model was trained on.
View project →A campus side-hustle platform connecting students who want short-term work with people on campus hiring for it. Its core idea is a refundable commitment fee: a student pays a small deposit to lock in a job, and it is returned alongside their pay once the work is approved, which cuts down on no-shows. Built on vanilla JS with Supabase for data and Stripe Checkout through serverless functions for payments.
Prototype complete — Phase 2, the student marketplace, is in planning.
View project →A hackathon concept for smarter waste handling: an AI chatbot that answers recycling questions, camera-based sorting to identify what an item is, and routing to the nearest bin using ultrasonic sensors to tell which bins are already full. Built in Flutter with Riverpod for state.
The interface and screen flow are built and the GPT chatbot is connected; camera detection and the sorting and routing logic are still ahead.
View project →A browser version of the fast, simultaneous card game, built to run as a Discord Embedded Activity so a group already sitting in a voice channel can start a round without leaving Discord. The rules engine — dealing, legal moves onto the shared centre piles, and scoring — is written as its own module, kept separate from the rendering.
Locally playable prototype with the rules and scoring working; real-time multiplayer over Discord is still to come.
View project →