The core claim behind Signpost is simple to state and harder to build: you sign, and within about a tenth of a second you know whether you got it right. This post walks through what happens in that tenth of a second.
From webcam frame to hand landmarks
Every frame from your webcam is passed to a hand-tracking model that returns a set of numerical landmarks: the coordinates of each joint in your hand. Nothing about this step leaves your device. The image itself is discarded immediately; what moves forward is a small array of numbers describing where your fingers are.
From landmarks to a graded sign
A sign in ASL is not a single pose. It is a handshape, a location relative to your body, an orientation, and often a movement over time. The model compares all of those against the target sign and produces a score, which is why a near-miss reads as a near-miss rather than silently passing.
Why latency is the whole game
Feedback that arrives a second late is not feedback, it is a grade. To correct a motor skill, the signal has to land while you can still feel what your hand was doing. That is the reason the entire pipeline runs locally instead of round-tripping to a server, and it is the reason the target is under 100 milliseconds rather than under a second.
What we store, and what we do not
Your webcam feed is never uploaded, recorded, or stored. The only data saved server-side is numerical landmark coordinates used to improve recognition, never images or video. More detail is in our security overview.
Stop practicing in the dark.
Signpost watches your hands through your webcam and tells you, in real time, whether your sign is right. Try the free demo, no sign-up needed.
Try the free demo