Work
VLM · Task camera
Camera step reader
A vision-language check on frames from a task camera: what step is this, and does it match the procedure.
The problem
Labeling every frame by hand does not scale for a small team. The model had to read a frame and say the step.
What we built
- Frames pulled from the working camera
- A vision-language call against the expected step
- Plain language back to the operator
- A queue for frames the model would not commit to
Tell us what you need built.
A short note on the device, the user, and the deadline is enough to start.