SignScope — Real-Time Sign Language Recognition & Learning Platform (AI training/data preparation for computer vision)
Data CollectionBuilt a real-time sign language recognition system that required collecting and organizing webcam-based training data for hand landmark–driven classification. Used MediaPipe hand landmarks to structure inputs and supported training/evaluation workflows for an alphabet recognizer. Integrated inference endpoints to produce model predictions during live sessions for continuous assessment during development. • Collected and prepared landmark-first sign language datasets from webcam sessions • Trained and evaluated a MobileNetV3 + BiGRU with attention word recognizer and an alphabet classifier • Implemented evaluation on a held-out landmark classification set with accuracy tracking • Deployed prediction serving via FastAPI for live learning feedback