$7M+ raised from Lightspeed Luminous, Factorial Funds and other top-tier VCs
Human data
for frontier AI.
Voice, conversation and private data from verified consumers and professionals, built by a team that runs its own human simulation research.
Trusted by builders and teams at
Supply
Name a profession.
We will find them.
Surgeons and school teachers, litigators and line cooks, pilots and plumbers. For a model, anyone who does a job well is an expert. We recruit by profession, seniority, market and language, and check credentials before the first task.
And hundreds more, from optometrists to orchestra musicians. How we source experts →
What they give you
Knowledge, taste and judgment.
Captured three ways.
On camera, in their own words.
Everyday people and experts each hold knowledge no one else has: what they prefer, why they chose it, how they would do it. They explain it to us on camera, in interviews and on tasks you design, and we deliver the recording with transcripts and profile context.
The real thing, recorded.
Plenty of work falls apart when someone acts it out for a camera. So, with everyone's consent, we record real sessions as they happen: a doctor with a patient, two lawyers on a contract, a mechanic mid-repair, a parent doing the laundry. Audio, video and screen, captured together.
Physical AI. First-person video from head-mounted cameras and smart glasses, hands and tools in frame, in real homes and workplaces.
An expert panel, on demand.
Some questions are hard enough that one expert is not enough. When you are pushing a model past the edge of a vertical, we convene several specialists to work the case together, argue it through and agree the best answer. You get the verdict, the reasoning and the disagreement along the way.
What we sell today
Five kinds of human data.
Three ways to get them.
Speech and conversation
Two-speaker and group recordings with overlap, pauses and barge-ins, plus labels for tone, emotion and delivery.
Learn more →Expert judgment
Doctors, lawyers, teachers, engineers and hundreds of other professions answering domain tasks and explaining their decisions.
Learn more →Private records
Consumer chats, AI histories, photos and purchases, and expert notes, contributed with consent and profile context.
Learn more →Enterprise data
Non-public emails, chats, documents, meetings and private codebases from consenting companies, sourced against a brief.
Learn more →Coding traces
Real developer-agent sessions from Cursor, Claude Code and Codex: prompt, tool calls, edits and outcomes.
Learn more →License
Existing recordings and records, with source context, metadata and usage rights.
Commission
New conversations, tasks and recordings from the people your brief describes.
Label and review
Experts annotate and grade against your rubric, with review notes.
Agree the spec and acceptance criteria → review a sample → scale.
Supply
Real people you can verify.
We work through 30,000+ local recruiting partners and a reachable network of 170M+ people. Identity, profession and background are checked before a task starts. Every session gets automated consistency checks when it ends, and trained annotators review what gets flagged.
For voice and interactive models
Your model can talk. Can it hold a real conversation?
Full-duplex and agentic voice models learn from people talking to each other: interrupting, pausing, saying "uh-huh", changing their minds. We record those conversations with consent and label every turn.
- Two-speaker and group recordings with overlap, backchannels and barge-ins
- Speaker turns, timestamps, tone, pauses and emotion labels
- Action timelines for agentic voice, from our Pull That Up research preview
- Many languages and accents, including low-resource ones, on request
Research
A lab that sells to labs.
We buy data, recruit people and run experiments for our own research in human simulation and evaluation. Other labs needed the same pipelines, so we opened them up. Data revenue funds the long-term work: models of real people, grounded in real interviews.
Human simulation
Model individuals from interviews and personal context, then compare predictions with what real people did.
Evals and benchmarks
Benchmarks judged by verified experts or everyday consumers, depending on who the answer is for.
RL environments
A virtual hospital, a law firm, a storefront, populated by virtual people grounded in real ones. Research, not yet a product.
Company
Built by people who have shipped AI at scale.
Based in San Francisco. $7M+ raised from Lightspeed Luminous, Liquid 2, Converge, UpHonest, Factorial Funds, GoAhead, Taihill.
Team and investors →Bring us your hardest data gap.
We'll run the sample.
Video on this site is licensed stock footage of real people, used to illustrate the kinds of data we collect.