Things I've built or am currently working on.
Cross-lingual system that recognises continuous sign language from video. Uses a 6-layer PyTorch Transformer with CTC loss. Covers German, American, and Indian Sign Language with 44K+ training videos. Deploying to Android, iOS, and Web.
PyTorch, Transformer, MediaPipe, TFLite, TF.js
GitHubEfficientNetB0 model for LSE (Lengua de Signos Española) with 95.73% accuracy across 19 classes. Uses Grad-CAM++ for explainability. Published at ICECMSN 2025 - first published work on non-English sign language recognition.
EfficientNetB0, Grad-CAM++, Transfer Learning, Python
Browser-based live demo - opens the camera via WebRTC, extracts pose with MediaPipe, runs a quantised TF.js model for real-time sign language inference. No install needed.
TypeScript, TF.js, WebRTC, MediaPipe
Java OOP system for hostel room management. Implements all four OOP pillars with 10 classes, file-based persistence, and full UML documentation.
Java, OOP, File I/O, UML
GitHub