Demand: AI teams
Bring a model need, budget, task spec, quality bar, and rights model. Humanbased turns that demand into a campaign the market can execute.
Not a catalog. A marketplace OS.
AI data is not inventory. It is production.
Humanbased is the marketplace OS for a category where supply often does not exist until demand arrives. Campaign Builder runs the data factory; Data Lineage turns accepted work into owned, attributed, distributable supply.
- Demand Demand creates the brief
- Production Campaign Builder governs the factory
- Lineage Data Lineage records attribution
- Distribution Value moves with the asset
Demand-side engine
Campaign Builder turns requests into data factories.
This is the factory floor of the marketplace. It defines objectives, task formats, contributor pools, agent assistance, review paths, quality gates, budgets, and acceptance rules so specialized AI data can be produced rather than merely sourced.
Airbnb has homes. Amazon has SKUs. AI data needs a factory first.
Traditional marketplaces begin with existing inventory. AI data marketplaces begin earlier: a model team knows what it needs, but the supply may not exist yet.
That supply has to be commissioned, generated, reviewed, attributed, packaged, licensed, and sometimes reused. Humanbased is built for that full loop: production first, marketplace second, lineage always attached.
The platform creates the asset between demand and supply.
Supply: contributors and experts
Complete accepted work, get paid primarily in stable-coin
payouts via
USDC
, build a portable contribution record, keep attribution, and
stay connected to future value when accepted work becomes a
reusable asset.
Operator supply: teams and organizations
Operate specialized supply groups with assignment, review, ownership, attribution, and payout rules that match each campaign.
Agent supply: automation builders
Plug pre-labeling, evaluation, QA, routing, auto-valuation, and campaign automation into the same production marketplace.
The operating loop that turns demand into distributable supply.
- Specify demand Define the model objective, task format, contributor profile, acceptance criteria, budget, and rights model.
- Run the factory Use Campaign Builder to route work through people, experts, teams, organizations, agents, review paths, and quality gates.
- Accept work Combine human review, expert judgment, agent checks, benchmarks, and provenance logs before the work becomes part of the asset.
- Record ownership Use Data Lineage to connect accepted work to attribution, provenance, ownership, and royalty eligibility.
- Distribute value Support upfront payment, service fees, access fees, ownership transfer, marketplace cuts, and future royalty distribution.
Supply-side foundation
Data Lineage turns production into ownable supply.
This is what makes the marketplace work after production. Once
work is accepted, Data Lineage records who contributed, what was
accepted, which rights attach, and who can participate when the
asset is distributed or reused. Humanbased can anchor ownership
onchain through
Base
and
Ethereum
so attribution, access, payouts, and royalty eligibility travel
with the data.
- Ownership records for accepted assets
- Attribution that survives assembly and reuse
- Distribution logic for payouts and royalties
Data Lineage is built on XnY so attribution, usage rights, and royalty paths stay with the data after it leaves Humanbased.
Specialized AI data starts as work, then becomes supply.
World modeling
Physical interaction traces, embodied task demonstrations, environment annotations, failure cases, and simulation feedback can move through the same production and lineage loop.
Expertise knowledge
Legal, coding, medical, financial, and scientific judgment often begins as expert review, annotation, reasoning, or validation work before becoming reusable model data.
Image and video
Consented media, scene descriptions, temporal consistency checks, safety review, preference ratings, and multimodal QA can retain provenance and rights context.
Voice
Licensed speech, consented recordings, accent coverage, dialogue review, pronunciation QA, and voice preference judgments can be produced with attribution attached.