Method · research compiled 15 Oct 2026
How the research is checked.
Each claim links to its evidence. Financial figures keep their dates and the part of the business they cover.
About and corrections
ScienceToStartup is published by Musa Ozdem. It is a public-source research guide to the companies that produce AI training data. No profiled company has paid for coverage, and none is a customer, investor or affiliate of the publisher. The publisher has applied for paid work at some profiled companies and is a member of A.Team's talent network.
Disclosure. The publisher is also proposing an AI research lab, described on the overview page. That proposal is separate from this research, which applies the same rules to every company.
Corrections. If a figure or a listing is wrong, request a correction with the company name and the source. We review each request within seven days.
Personal data and removal. Company profiles name founders and executives from public company sources only. Profiles do not show personal contact details. If you are named here and want a record corrected or removed, send a correction request. Visits are counted by Cloudflare Web Analytics, which sets no cookies.
How figures are shown
- Each figure keeps its own date and the part of the business it covers, such as the whole company or one business segment.
- Key figures come only from supported claims about reported results. The interactive guide uses the same rule.
- Figures from talks, unverified reports or conflicting sources appear only in a separate list marked Not confirmed, with their status in words.
- Historical marks figures from 2024 or earlier.
- Annualized pace is recent revenue projected across a full year. We say whether revenue is gross or net when the source says so.
- Estimates are labeled. Forecasts, targets and older or partial figures go in a separate list of other reported figures. They never count as key figures.
- Later source corrections appear next to the figure they correct. The original records stay in the downloads.
Confidence grades
Each claim carries a confidence grade that says how strong its sources are.
- A
- Independent or buyer-side evidence, such as a filing or a customer who confirms the work.
- B
- The company, an investor or another named first-hand source says it.
- C
- Reporting by others, an estimate, or indirect evidence.
- U
- Not confirmed. The fact may be undisclosed or still unresolved.
What the labels mean
Each company also gets a delivery evidence level (public proof that it sells and delivers its offer). The level comes from four measures, each rated from level 1 to level 3, and covers one part of the business, such as the whole company or its AI data work. Funding and valuations do not count toward it.
| Measure | Level 1 | Level 2 | Level 3 |
|---|---|---|---|
| P · Product / delivered offer | Concrete current offer and identifiable use case | Accessible product documentation, demo, reproducible artifact, or detailed delivery workflow, with dated activity | Level 2 plus a documented production deployment and continuing product/service releases |
| C · Customer / adoption evidence | Supplier claims customers, displays logos, or gives an anonymous aggregate | At least one named deployment described in a detailed case study or attributable customer testimony | At least one named deployment independently corroborated by the buyer, a filing, procurement record, or other direct transaction-side evidence |
| R · Repeatability / commercial continuity | One dated paid project, contract award, revenue disclosure, or explicit paying-customer claim | At least one renewal/expansion, or two comparable dated revenue periods, or two distinct dated commercial deliveries at least six months apart | Level 2 plus at least three distinct named buyer organizations and evidence of commercial continuity across at least 24 months; alternatively, financial statements establish revenue across two annual periods |
| O · Operating readiness / delivery discipline | Named accountable team and documented delivery process | Level 1 plus specific QA/acceptance methodology and buyer-facing operational/security documentation | Level 2 plus independently evidenced operational performance or independently assessed controls with current scope and dates |
Research coverage
39 company profiles · 37 companies with a delivery evidence level · 880 source records. The two nonprofit research groups get no level.
194 claims are unverified or disputed. Each one is labeled. The guide is a selected study of the market. It does not list every company.
Download the data
Dataset version 2.5.0, research compiled 15 Oct 2026. The research data files keep every claim that later research replaced.
ScienceToStartup licenses this data under the CC BY 4.0 license (any use is allowed if you credit us). Cite ScienceToStartup and the original sources. Company names and logos belong to their owners. The license does not cover them. It also leaves out product images, photos of people, text quoted from sources and personal contact details.
| File | Format | Size | Contents |
|---|---|---|---|
| research.json | JSON | 3.2 MB | The core research data in one file |
| observatory.sqlite | SQLite | 9.4 MB | The same data as one database file |
| work-routes.json | JSON | 455 kB | Paid work listings and their sources |
| source-updates.json | JSON | 106.3 kB | Later source corrections and the values they replace |
| profile-content.json | JSON | 163.2 kB | Profile summaries and the records they cite |
| editorial.json | JSON | 31 kB | Business groups and plain-language summaries |
| business-health.json | JSON | 194 kB | Each business health dashboard observation, with its sources and calculations |
| assessments.csv | CSV | 55.1 kB | The assessments table |
| claims.csv | CSV | 1.2 MB | The claims table |
| classifications.csv | CSV | 11.2 kB | The classifications table |
| companies.csv | CSV | 19.9 kB | The companies table |
| company_investors.csv | CSV | 3.5 kB | The company investors table |
| company_people.csv | CSV | 12.7 kB | The company people table |
| company_relationships.csv | CSV | 840 B | The company relationships table |
| customer_evidence.csv | CSV | 40.5 kB | The customer evidence table |
| editorial_readouts.csv | CSV | 19.1 kB | The editorial readouts table |
| funding_rounds.csv | CSV | 34.9 kB | The funding rounds table |
| investors.csv | CSV | 12.1 kB | The investors table |
| metrics.csv | CSV | 113.5 kB | The metrics table |
| people.csv | CSV | 3.9 kB | The people table |
| products.csv | CSV | 65.4 kB | The products table |
| round_investors.csv | CSV | 52.6 kB | The round investors table |
| sources.csv | CSV | 249.2 kB | The sources table |