First-person video with both hands tracked, 21 keypoints each, every frame.
// physical data for ai
the future is
in your hands.
Farasa records how people work in the physical world: what they see, how their hands move, what they touch, and what the machines beside them do. Every frame labelled, so robots can learn real work.
// 01
Every sentence ever typed became training data. Almost none of the work done with hands did. That’s where we come in.
// 02
what we capture
Telemetry from the machines on the same job, from position and load to speed, synced to the video on one clock.
Objects, contact, and actions, labelled frame by frame by trained annotators, then two rounds of QA.
// 03
how it works
- 01
scope
We agree on the tasks, sensors, and label schema with your team.
- 02
capture
People do the real work while we record it from where they stand, with every sensor synced.
- 03
label
Trained annotators label every frame against your spec, then two rounds of QA.
- 04
deliver
The dataset ships in your format, ready to train on.
// 04
custom datasets
Need something we haven’t recorded yet?
Physical data is our focus, but we also build custom datasets to spec: new tasks, new sensors, new label schemas, at the scale you need. Tell us what your model is missing.
the internet taught machines to talk.nothing has taught them to work.