What we do
Saolabs sources pre-vetted experts in medicine, law, finance, engineering, software, science, mathematics and languages. They produce expert-labelled data for AI model companies and labs: labelling, training data, evaluation, red-teaming and alignment data.
The work spans language, code, vision, voice, video, agents, simulation, robotics and the physical world. We do not build our own model. We help make other people's models and agents safer.
Why we exist
A model learns what it is shown. If the people grading its answers cannot tell a sound diagnosis from a plausible one, or a valid argument from a confident one, the model learns the wrong lesson.
So the safety of an AI system depends, in large part, on the expertise of the humans who teach and test it. We started Saolabs to take that dependency seriously: find people who genuinely know their field, check that they do, and put their judgement where it matters.
How we work
Every expert is sourced, vetted and matched. Sourcing runs through our agentic Hiring OS, outreach agents, a recruiter network and a jobs marketplace. Vetting combines AI voice interviews, real-world work tests and skills assessments.
Only then are experts matched to projects that fit what they have shown they can do.
The technology behind it
To vet experts at scale we had to build our own hiring technology. It became a complete ecosystem: Hiring OS, a jobs marketplace, AI voice agents, outreach agents and Agent OS.
We now license it to other companies, for hiring and for running their own workflows on agents trained with expert human guidance.
The name
Sao is a small moon of Neptune. It was discovered in 2002, named after one of the Nereids, and travels on a steeply tilted orbit.
We liked what that suggests: a small, independent body doing serious work on its own trajectory.