One person, before the team, the tools, the clients. Kept going until there was a company. Has not stopped — and is not planning to.
Six zeros is what turns one into a million. That is the work in between — and the only part that actually matters.
A million people earning through this platform. Not one-off tasks — steady, remote, human work.
That is the name. The rest is the work of earning it.
There was someone who liked helping people — any kind of helping. That instinct narrowed into one question: how do people find real work?
AI cannot train without human input. That input is needed forever, from as many people as can do it well.
We take that demand and make it a paycheck. Remote by default. Paid on time. Every week.
Multilingual speech capture, transcription, and labeling.
Writing, reviewing, and rating programmer output.
Step-by-step reasoning and proof grading.
Translation, annotation, and multilingual data.
Comparing and ranking model responses. RLHF at the seam.
It starts with a brief from you. It ends with a batch delivered back to you, with a score printed on it that shows how many items missed the mark. In between, your work moves through seven steps. A standard we both sign. A pool trained against it. A first read. A second read. A final call on the edges. A quiet layer that catches drift. And a published miss rate that ships with every batch. No model touches your data. Every step happens in our environment, with people who have proven they can read your standard. Follow it, step by step.
Before a single item is produced, your brief becomes a one-page standard we both sign.
We read your brief, ask what you actually mean by good, and write it back as a one-page rubric any reviewer can act on. It covers examples, edge cases, and a no-go list. You sign it before production. When a case resolves that changes the rule, the page is reissued, and the changelog ships attached to your next delivery.
A dedicated pool per category, cleared on your standard.
Every category runs its own closed pool: voice, code, math, language, evaluation. Every reviewer in that pool has cleared a live trial against the same standard we signed with you. The trial is real work, not a quiz. Only cleared reviewers are assigned your items.
Every item is graded against your rubric by a trained reviewer, in our environment, with no model involved.
No automation sees your data. Every item is graded in our environment, on our infrastructure, by a reviewer who has cleared the trial for this category. The rubric is tight enough that grading is decisive, not debate. Clean items move toward delivery. Doubtful ones, plus a sample of the rest, move to a second read.
Anything doubtful, plus a sample of the rest, reaches a senior reviewer. Review effort follows risk.
A senior reviewer sees everything flagged on the first read, plus a random sample of clean items to keep the first read honest, plus more of both for any producer whose track record is new or drifting. The cost is light where the risk is light, and concentrated where it is not.
Real edges escalate to whoever wrote the rule with you. Their decision ships, and it updates the standard.
A small share of items, the genuine edges, do not resolve on the second read. They escalate to one of the people who wrote the standard with you. Their call ships. It also rewrites the relevant line on the standard, so the same case never gets argued twice.
Items we have already graded ourselves sit in every queue. Drift gets caught the same day.
Known-answer items sit in the queue looking like any other task. A reviewer cannot tell them apart. They tell us, in real time, whether a reviewer's reading of the standard is drifting. We use them to catch problems the day they happen, before they reach your batch.
Every delivery publishes its miss rate against the standard. Per batch. Not averaged.
Before a batch leaves us, we audit a slice against the same signed standard. Items that miss are returned, redone, and resolved inside the same cycle. The miss rate ships alongside the delivery, with the items that missed and why. You never get a quiet pass.
Reviewers, writers, domain experts. You read carefully and you want that to count. We hire for judgment and we pay for it.
Start application →Reasoning, code, writing, domain work. You bring the rubric and the edge cases. We deliver the work, graded, with the misses in the envelope.
Request a conversation →Reviewers, writers, domain experts. You read carefully and you want that to count. We hire for judgment and we pay for it.
Reasoning, code, writing, domain work. You bring the rubric and the edge cases. We deliver the work, graded, with the misses in the envelope.