Backed by Y Combinator

A fast, accurate video decision model.

Ask Nua to find any moment in hours of footage. It pinpoints exactly where it happens and makes the call: did it happen, for how long, and did it pass.

teleop_episode_118.mp40:00 / 1:34

Training data labeling

“Did the robot put the mug in the bin?”

0:001:34
site_cam_dock_2.mp40:00:00 / 6:00:00

General liability and workers' comp claims

“Does the worker lift the box off the pallet, and when?”

0:00:006:00:00
partner_stream_q3.mp40:00:00 / 2:04:00

Co-op and sponsorship verification

“Is the Halcyon logo on screen for at least 8 minutes?”

Yes.9m 55s on screen across 4 placements
0:00:002:04:00

Your footage goes here.

That's how teams are using Nua right now. Tell us what you need found, and we'll run it on your video with you.

Try:
0:00:00your_footage.mp4
Benchmarks

Artificer task evaluation

Artificer scores 92.84% balanced accuracy on ArmNet268 and 0.4308 source-averaged mean absolute error on RoboReward.

ArmNet268: balanced accuracy (higher is better). RoboReward: source-averaged mean absolute error (lower is better).

Benchmark results

Scroll horizontally to view both benchmarks.

Read the Artificer launch and evaluation details →

Bring us your use case.

We're working with a small number of teams on real video and actively looking for new problems to point Nua at. Bring a sample and the question you need answered, and we'll show you what it finds.