AMERICAN SIGN LANGUAGE · AI WORKBENCH
Conversation, seen differently.
A focused workspace for hand tracking, sign practice, and model-backed recognition.
On-device visionBrowser processed
0 samplesLocal dataset
Bidirectional workspace
Live capture
Real-time hand landmark tracking
Ready when you arePosition one hand in frame to start tracking landmarks.
Landmark detector idle
Capture training sample
Save the current normalized 21-point hand shape
Only landmark coordinates and your label are saved locally. No image or video is recorded.
Recognition output
Classifier predictions appear here
LATEST PREDICTIONWaiting for a trained model
Hand tracking is active, but no ASL classifier is configured yet.
Connect local model APISends landmark vectors only, never camera frames.
Session transcript
Build a phrase from reviewed predictions
Your accepted predictions will appear here.
Recognition vs. tracking A visible hand skeleton means the camera pipeline works. It does not mean an ASL sign has been identified.