We collect, process and trade AI training data.
Buyers bring a specification: the data type, the languages, the volume, the quality bar. We collect the data with our own field and studio teams, process it in-house, or license it from producers we have vetted, and deliver it as a documented dataset. Nothing is held in stock.
- Data categories
- 36
- Languages
- 120
- Language × category combinations
- 468
- Writing systems
- 39
What we do
- Collect. Our own field and studio capacity, plus a vetted producer network for specifications we cannot meet in-house.
- Process. Transcription, annotation, quality control and packaging run in-house, against a guideline agreed before collection starts.
- Trade. We license and deliver datasets, including ones we did not collect, where the licensing chain can be evidenced.
How we work
- A written specification comes first. Data type, volume, contributor count, language and dialect, collection environment, annotation depth, delivery format. We do not quote against a vague request.
- We check before we quote. Producers are matched against the specification, not against a category label. If nobody in the network can meet it, we say so rather than take the order and improvise.
- A pilot batch is reviewed before the full run. The cheapest way to catch a specification gap: hours of work instead of hundreds.
- Delivery includes the documentation. Data files, labels or transcription, contributor metadata, the annotation guideline, consent records and the quality report, as a set.
Quality and acceptance
Acceptance criteria are agreed before production, not argued about after delivery: sampling rates and channel layout, transcript accuracy against a held-back sample, label agreement between annotators, duplicate rates, contributor spread. When a batch fails the agreed bar we find the cause and re-run it. Re-work caused by a guideline we wrote or a producer we selected is ours to absorb.
What we do not do
These are jobs we turn down, including paid ones.
- No undocumented resale. We license and trade existing datasets, but only where the chain can be evidenced and the rights reach you.
- No medical or clinical data. The consent and privacy exposure is not something we are set up to carry.
- No recorded telephone calls. The same consent problem, and no way to evidence it at scale.
- No unprovable provenance. If the consent chain cannot be shown, the data does not ship, however good it is.
Work with us
-
If you are buying data
Send a specification and we will tell you whether it is sourceable, what it will cost and how long it will take.
-
If you produce data
Studios, annotation teams and field operations: the supplier programme is where to start. We are actively looking for capacity in low-resource languages and embodied AI.
Why a broker, in more detail · Compliance, consent and licensing · How a project runs
Tell us what you need
Tell us the language, the hours, and what the data needs to look like. You will get a real number and a real timeline, not a range. If we cannot source it well, we will tell you that instead.
- Pilot batch before the full run, so problems surface early.
- Consent documentation delivered with the data.
- No medical or clinical data. No recorded telephone calls.
- You send the spec Language, hours, recording conditions, and the deadline.
- We reply within two business days A real number and a real timeline, not a range.
- You approve a pilot batch A small run first, so problems surface before the full order.
- Full run, then delivery with documents Audio, transcripts, and the consent record for every speaker.