About
Public web data has been consumed. The next gains come from records that exist only inside institutions: chart notes, code reviews, audit trails, recordings of people doing their jobs. DataOrigin exists to license that data lawfully and make it usable for training and evaluation.
Illustrative, indexed. Training demand keeps compounding; the usable public corpus does not.
01 · Why we exist
Premise 01 · Supply
Every lab can buy the same chips and read the same web. What separates models now is the distribution they were trained on, and the most valuable distributions, how medicine, engineering, finance and operations are actually done, sit in institutional archives that have never been licensed.
Premise 02 · Provenance
We license from the owner, log every transformation, and write scope, term and revocation into the contract, so that for any record a lab can answer two questions: where did this come from, and are we permitted to use it?
Premise 03 · Judgement
The right answer in a chart note or a code review is not a majority vote. We admit under 2% of applicants, confirm credentials with the issuer, and re-calibrate every expert weekly, because a rubric is only as good as the person who wrote it.
02 · Operating principles
Every record enters the catalog under a licence signed by the institution that owns it. If we cannot name the source, it is not delivered.
Scope, exclusivity, term and revocation are decided by the data owner and enforced after delivery. We do not resell outside the agreed scope.
Identifiers are removed and every redaction is logged. Residual risk is reviewed by a person, and 42% of ingested records are rejected.
Trials are paid, work is paid on time, and re-calibration is transparent. People at the top of their field do not work for exposure.
03 · Where we sit
Institutions hold the data and cannot train on it. Labs can train and cannot reach the data. DataOrigin is the layer between them, and stays independent of both.
Hospitals, enterprises, studios and engineering orgs that generate valuable data as a by-product of their work.
Licensing, de-identification, normalisation, practitioner review, provenance and packaging.
Working replicas of production systems, real tasks and practitioner grading, versioned together.
Labs, enterprise AI teams and sovereign programmes that train, evaluate and deploy models.
04 · The company
DataOrigin is an independent data infrastructure company. We do not train models and we do not compete with the institutions we license from. That is what lets both sides trust us with the middle: institutions license records they have never released, and labs receive datasets with provenance they can audit. The expert network is global and remote, staffed by practitioners who are verified, examined and paid, wherever they work.
We are pre-launch. Figures across this site are targets, not audited results, and no customer, partner or certification is implied.
For AI labs and enterprise teams
Tell us the capability gap. If the catalog covers it, you will have samples under NDA within days.
Request samplesFor data owners
License it on your terms. We handle de-identification, compliance and delivery; you keep control of scope, term and exclusivity.
License your data