Human data for robotics and AI
Real-world human data, collected to your spec.
We recruit, train and manage contributor teams that record egocentric video, speech and workflow demonstrations. Every batch passes two layers of quality control before it reaches you.
What we collect
Four kinds of work, one managed team.
We are not tied to one data type or country. We recruit for each approved project where you need people.
Egocentric and workplace video
First-person recordings of real tasks at home and at work: kitchens, workshops, warehouses, repair and assembly.
phone or head-mounted camera · your devices welcome
Speech and conversation
Scripted prompts, natural dialogue and wake-word recordings from native speakers, with speaker metadata.
48 kHz WAV · SNR and clipping checked
Desktop workflow recordings
Screen recordings of real software tasks, captured step by step with action logs.
screen + input events · PII masked
Annotation and review
Action segmentation, transcription, labelling and quality review on your platform or ours.
gold tasks · per-reviewer accuracy
How a project runs
Start small. Scale on your acceptance rate.
Acceptance and payment per batch, so problems surface early instead of at the final invoice.
- Brief and specYou share the task, capture protocol and acceptance rules. We ask until nothing is ambiguous.
- SamplesWe record 2–3 samples to your spec so you can confirm we read it correctly.
- Recruit and onboardWe recruit contributors for the countries and profiles you approve, train them with step-by-step guides and a test clip.
- CaptureRecording with automatic checks on the device: framing, lighting, duration, audio.
- Two-layer QAAutomated checks, then human review of every batch before delivery.
- Batch deliveryFiles and metadata in the format your pipeline expects, with a rejection report you can audit.
Quality control
What every file is checked for.
Built from the rejection reasons data buyers report most often.
Video
- Resolution, frame rate and orientation
- Hands visible for the share you require
- Lighting, focus, shake and obstructions
- Task performed as briefed, no idle stretches
- No speech, music or bystanders where excluded
- Duplicate and near-duplicate detection
Audio
- Signal-to-noise ratio and clipping
- Loudness within target range
- Long pauses and silence
- Echo and second voices
- Read-back accuracy against the script
- Format, sample rate and metadata
Contributors are identity-checked, limited to one account each, and given clear reasons with every rejection so the next take is better.
Consent and rights
Clean provenance for every asset.
- Signed consent from each contributor before recording, versioned and stored with the asset.
- Workplace releases from the owner of every recording location.
- No minors and no bystanders without consent; faces and personal data masked on request.
- Rights you choose: exclusive or non-exclusive use, agreed per project and recorded per asset.
- Data handling designed around GDPR principles, with deletion on request.
Who we are
An early-stage team, built per project.
- We recruit for each approved project rather than claiming a large standing workforce.
- We start with samples and a small paid pilot, then scale on your acceptance metrics.
- We work with your hardware, apps and annotation tools, or provide our own workflow where you prefer.
Contact
Send us a brief.
Tell us the task, countries and volume you need. We reply with questions, a sample plan and pilot terms.