Humanbased

Launch AI data campaigns as programmable workflows.

Humanbased gives AI builders one place to commission data, train contributors, use agents, review quality, and turn accepted work into reusable assets.

Campaigns are running now in invite-only beta. For a guided pilot, write to hello@humanbased.ai.

Read the docs Start a campaign

The campaign loop

  1. Design Define task schemas, rubrics, budgets, rights, and acceptance paths.
  2. Route Match work to contributors, experts, teams, agents, and validators.
  3. Verify Combine automated QA, human review, expert judgment, and provenance.
  4. Reuse Package accepted outputs as datasets, evals, feedback corpora, or data products.

The loop as a canvas

Campaign Builder is where those four steps become one editable graph. Each node carries its own instructions, and the nodes that route work to people also carry a contributor pool, agent assistance, and a per-record price, so the budget and the quality gates are visible while you are still designing the campaign.

Campaign Builder canvas for a night-time street-sign campaign, inside the developer portal: an import step, an entry gate set to CRS ≥ 60, two paid human steps priced per record in USDC, deduplication, and peer review with a rework loop — beside an Agent Mode panel confirming the entry gate, a contributor task, and an acceptance point are all present.

What you run it with

Campaign Builder

Model a campaign as a workflow instead of a spreadsheet. Each node owns instructions, inputs, contributors, agents, review rules, and downstream asset assembly.

Agentic workflow

Agents handle pre-labeling, routing, rubric checks, retrieval, scoring, validation, and packaging while humans keep judgment, consent, edge cases, and final acceptance.

Academy

Training that qualifies contributors before they enter a campaign: lessons, practice tasks, quizzes, calibration, and versioned completion rules.

Marketplace assets

Accepted campaign output becomes reusable assets with provenance, license terms, API access, ownership records, and royalty routing.

What you control

You set the task schema, the rubric, the contributor gates, the budget, and the acceptance rules. Nothing about the pipeline is a black box: routing, review depth, and agent assistance are choices you make per campaign.

Quality is a tuning loop, not a promise. Teams adjust qualification gates, cohort mix, rewards, gold sets, and validation until cost, data quality, and volume sit at the right operating point.

Go deeper

The full operating model Compare the alternatives How it works for contributors