BSL Sign Gloves is a wearable system that recognizes British Sign Language (BSL) two-handed fingerspelling (the letters A–Z) in real time and assembles them into on-screen text. 

Two ESP32-powered gloves — one for each hand — stream synchronized sensor packets (finger bend, fingertip pressure, and hand orientation) over Wi-Fi at 50 Hz to a FastAPI backend. The backend aligns the left and right hand frames by timestamp, engineers a 26-dimensional feature vector, and routes the stream through a two-stage machine-learning recognizer: 

a static classifier (XGBoost, with Random Forest and SVM baselines) for the 24 still-pose letters, and 
a dynamic CNN-LSTM for the two motion letters H and J (plus a set of distractor gesture classes that teach the model to ignore everyday movement). 
Recognized letters are debounced, displayed on a React dashboard, and accumulated into words and sentences. 

💡 BSL uses a two-handed manual alphabet (unlike ASL), so both gloves are mandatory — one hand frequently acts on the other, and the inter-hand relationship is part of the signal. 

The project is intentionally synthetic-data-first: because no public BSL sensor-glove dataset exists, the entire ML pipeline is developed and validated on synthetic data generated from BSL SignBank sign definitions, then progressively fine-tuned on real captured data