Raw first‑person video in. A dataset you can sign for, out.
Deepen Refinery checks every clip against your order, removes faces and screens, transcribes, and narrates with AI or with people, then delivers with a manifest you can verify.
Five things that used to be five vendors.
One order, written in your words, runs the whole line. Pick a station to see what it does to a clip.
Every clip, checked against your order.
Resolution, length, duplicates, sound, how much of the clip is narrated, whether the camera is worn by the person doing the work. Your floors, your units. A clip that fails is turned away with the reason; a clip the model is unsure about waits for a person.
Collectors see the same reason you do, in plain words, with a list of what to replace.
Faces and screens gone before anything leaves.
Faces are found and blurred on every frame. Phone screens and documents are caught and, if legible, the clip is turned away. A box the detector is unsure about is decided by a person, not delivered on a guess.
A delivery is held until every clip going out has been through this step.
Word-level transcripts you can search.
Every clip is transcribed with timings per word, so a dataset can be searched by what was said and when. Speech that is not there is not invented: a silent clip is turned away, never given a transcript.
Transcripts ship beside the clip in the format your order names.
0:00I am soldering component onto the green circuit board carefully. I set down my soldering iron and adjust my sitting position.
0:11I pick up the soldering iron again to continue working. I apply solder to the joint and hold it steady.
Narration by AI, by people, or both.
An AI narrator writes first-person, present-tense narration from the frames. A human narrator can record it, or record their own, and every take is checked by machine and then by a person before it ships.
You choose per order: AI script, AI voice, human voice, or human narration end to end.
A teleprompter built for the picture.
Narrators record against the video, segment by segment, with the script paced to what is on screen. A segment that misses is re-recorded on its own, not the whole clip.
Works in the browser with a microphone; nothing to install.
A portal for each side of the order.
Buyers see what they were sent and what they are billed for. Suppliers see what was accepted and what to replace. Nobody has to ask.
For the team buying the data
Your deliveries with their manifests, any clip reviewable by name, and a ledger where what you are billed is what you accepted.
For the people supplying it
Upload in tranches, watch them taken in, and get a plain reason and a replacement list for anything that did not make it.
Built by Deepen AI
Deepen Refinery is built and run by Deepen AI, a team that has spent years building annotation and data infrastructure for physical-AI programs. Refinery brings that to first-person video: the order in your words, every clip accounted for, a manifest you can verify.
Send footage. Sign for a dataset.
Tell us what the footage is and what the dataset has to be. We write the order back in your words and run the line. Already working with Deepen? Sign in to your portal.