01
No invented proof
We don't show client logos, testimonials or partners we haven't signed. Our evidence is published research you can check yourself.
About
Expertise AI can't scrape, verified before it trains.
Frontier models learn from what is online. The judgement of a clinician, an aerospace engineer, a patent attorney or a native speaker mostly never made it there. We find the people who hold that expertise, anywhere in the world, verify them, and turn what they know into data AI labs can trust.
What we do
We are building a global, multi-domain expert workforce for frontier AI. We recruit and vet physicians, pharmacologists, lawyers, scientists, engineers, defence and policy analysts, educators, finance professionals, artists and native speakers of more than 100 languages. Every person passes our vetting before touching client work. They then produce RLHF and preference data, evaluations and red-teaming, reasoning rubrics, domain Q&A, multimodal annotation and transcription for AI labs, and we pay them in US dollars.
Why vetting is the product
An AI lab buying human data is buying trust: that the person rating a medical answer is a doctor, that the physics rubric was written by a physicist, and that the translation was checked by a native speaker. Our five-check process, from identity to continuous quality, is recorded for every expert so clients can rely on it.
Our method, in public
Our first published evaluation is one example of the method we apply in every domain: truth written blind by a vetted expert, adjudicated against the source, and scored item by item. On a historical Urdu Nastaliq sample, a frontier model made at least one mistake on 114 of 161 lines, and left 55 of its 84 reading errors unflagged. See the method and results.
Who we are
Vettedlance recruits, verifies and manages domain experts for AI teams, and works with experts worldwide.
What we hold ourselves to
Trust is the product, so we don't fake it.
02
Verification before volume
Every expert passes identity, credential, assessment, calibration and ongoing quality checks before and during client work.
03
Provenance by default
Consent recorded for every contributor and a chain of custody for every item, exportable with each delivery.
04
Fair work
A written contract for every contributor, a rate agreed before each project, payouts on a published schedule, and no fee to apply.
Proof, published
We gave AI models a 1908 Urdu text. The best still got 1 word in 11 wrong.
A vetted native expert typed the truth blind, every disagreement was checked against the scan, and every word was scored. On our public leaderboard the top model, Claude Fable 5.1, reproduced 91.08% of words exactly and still made a mistake on 111 of 161 lines. Many errors carry no warning flag, so nothing in the output tells a reviewer to look twice.
- 91.08%
- best strict word accuracy (Claude Fable 5.1)
- 111/161
- lines with a mistake, even for the best model
- 2,378
- words and marks typed blind by a vetted native expert
- 8
- models ranked so far, more added as they finish
The same method runs in any field: truth written blind by a vetted expert, adjudicated against the source, scored item by item.
Get started
Put verified expertise behind your model.
Start with a scoped pilot, or join the expert network. Either way, verification comes first.
For AI labs & data vendors
Scope a pilot in your domain, languages and quality bar. We reply within two business days.
For experts
Paid, remote AI work in your field. Free to apply, with a written contract for every project.