Provenance-first data licensing · Est. on 13 years of data

The origin layer for frontier AI.

Proprietary, provenance-tracked data — created by identified experts under documented consent — licensed to the labs, enterprises, and sovereign programs building models that have to be right.

Request data samples → Become a data partner
0
curriculum-aligned questions
0
expert video
0
lines of production code
0
operational data
0
egocentric video — and counting
A two-sided data platform

Data with a verifiable past,
licensed for the long run.

One principle on both sides of the platform: data used for ongoing training deserves ongoing compensation — and a chain of custody you can show your lawyers.

01 — For Model Builders

License rights-cleared, AI-ready data by vertical.

  • Curated assets across education, STEM video, egocentric video, code, and operational workflows
  • Request samples before any commitment
  • Ongoing-supply and quality-certification options — never one-time dumps
  • Full provenance documentation with every delivery
Browse data assets →
02 — For Data Providers

Turn proprietary data into recurring revenue.

  • Ethical, governed licensing of underutilized data assets
  • We handle provenance, IP protection, and compliance
  • Recurring compensation when your data trains models on an ongoing basis
  • You control who licenses what, and for how long
Become a data partner →
Data assets

Organized by vertical.
Verified at the source.

Every asset is rights-cleared, provenance-tracked, and ready for training or evaluation — governed under documented consent and compliance review.

EducationAvailable now

Multimodal K-12

4M+ curriculum-aligned questions across 12 Indian languages, 33 boards, Classes 1–12 — tagged by board, class, subject, chapter, topic, and difficulty (L1–L10), with step-by-step solutions by 400+ subject-matter experts. Plus 3.8M expert video walkthroughs, 5,100 hours of lectures, and 4,500 hours of mentor audio.

4M+ questions12 languages 33 boards3.8M walkthroughs

Best for: vernacular reasoning benchmarks, hallucination reduction, difficulty-calibrated evaluation.

STEM VideoAvailable now

STEM Video Library

~80,000 PhD-level lecture videos — roughly 40,000 hours of native 1080p footage, primarily English-language sciences. Deep, structured exposition of advanced STEM at a scale the open web can't assemble.

~80,000 videos~40,000 hoursNative 1080p

Best for: multimodal pretraining, scientific reasoning, video-language alignment.

Egocentric VideoAvailable now

First-Person POV Corpus

9,200+ first-person videos spanning 30+ rarely captured domains — tech repair, cloud kitchens, gaming, skiing, restoration, craft and industrial work. The value is access: domains the open web doesn't cover, recorded under documented consent with full provenance.

9,200+ videos30+ domainsConsent-documented

Best for: embodied AI, robotics perception, multimodal action understanding.

EngineeringAvailable now

Production Codebase

16.8M lines of production code across 95 repositories with 13 years of continuous git history — JavaScript, PHP, TypeScript, and Python. Real engineering decisions, refactors, and fixes. Not synthetic snippets.

16.8M lines95 repos13 yrs history

Best for: coding-agent training, eval benchmarks, RLHF on real-world code.

OperationsAvailable · NDA

Operational Workflow Data

Anonymized enterprise communication and CRM workflow data capturing how cross-functional teams actually decide — escalations, handoffs, approvals, resolutions over time. Shared under NDA.

AnonymizedCRM + commsNDA required

Best for: enterprise & agentic AI, workflow modeling, decision-pattern learning.

HealthcareAvailable · NDA

Healthcare Data

De-identified clinical and health-operational data, governed end to end — anonymization and consent verified before any delivery, under full data-protection review. Shared under NDA.

De-identifiedConsent-verified DPDP-compliantNDA required

Best for: clinical reasoning, medical Q&A, healthcare workflow modeling.

Voice & AudioAvailable now

Conversational Audio

Large-scale real-world voice conversations — natural, multi-speaker, real-context audio rather than studio recordings. Prepared under documented consent and DPDP-aware privacy processing.

Multi-speakerReal-context Consent-documentedDPDP-aware

Best for: speech recognition, speaker diarization, conversational & voice agents.

Why DataOrigin

Hard to replicate.
Easy to defend.

Anyone can scrape. Almost no one can show where their data came from, who made it, and that they had the right to license it. We can — for every asset, every time.

⌖

Full provenance

Every asset traces to identified creators, with documented chain of custody from creation to delivery.

✓

Documented consent

Content created by experts under explicit, recorded consent — no ambiguity about rights or downstream use.

अ

Multilingual depth

Twelve Indian languages with curriculum-grade tagging — vernacular depth the open web simply doesn't have.

§

Governance & compliance

NDAs, governed delivery, anonymization where required, and DPDP-aware processing throughout.

∞

Ongoing supply

Pipelines that keep producing — refreshed, expanded, and quality-certified across the life of a license.

◆

Recurring licensing

If data trains models on an ongoing basis, the holder is compensated on an ongoing basis. That's the model.

provenance — verify
How it works

From sample to ongoing license.

A governed path designed for legal and procurement teams as much as research teams.

Sample review

Request representative samples of any asset. Evaluate quality, coverage, and fit on your own benchmarks.

NDA

Mutual NDA covering deeper access, documentation, and any asset shared under restricted terms.

Commercial terms

Scope, exclusivity, refresh cadence, and recurring-licensing structure agreed up front.

Governed delivery

Secure, audited delivery with provenance documentation attached to every dataset shipped.

Ongoing licensing

Continuous supply, quality certification, and renewals — a relationship, not a transaction.

Who it's for

Matched to what you're building.

Sovereign & national AI

Multilingual education + STEM video. Models that reason in twelve Indian languages and command advanced science — with data your program can publicly stand behind.

→

Enterprise & agentic AI

Production codebase + operational workflow data. Train agents on how real organizations write software and make cross-functional decisions.

→

Coding agents

Engineering codebase. 13 years of real commits, refactors, and fixes — the longitudinal signal synthetic data can't fake.

→

Multimodal, embodied & robotics

Egocentric + STEM video. First-person footage from 30+ rare domains plus 40,000 hours of structured expert exposition.

→

EdTech

Education corpus. Curriculum-aligned, board-verified, difficulty-calibrated — ready for tutoring models and assessment engines.

→
For data providers

Your data is an asset.
We make it behave like one.

Most organizations sit on proprietary data AI labs would pay for — but lack the provenance documentation, legal scaffolding, and buyer relationships to license it safely. That's exactly what we built.

Become a data partner →
13 years of data history 400+ subject-matter experts 33 education boards Institution-verified Documented consent Anonymized & de-identified DPDP-aware processing Compliance-first Compliance badge — coming
Contact

Start with a sample.

Tell us what you're building or what data you hold. We'll respond with relevant samples, documentation, or next steps — usually within two business days.

partnerships@dataorigin.ai