Centaur

Centaur.ai is a data annotation platform focused on high stakes domains, especially healthcare AI. It produces labeled training and evaluation datasets by having human experts and AI models annotate the same items independently and then combining the best signals. The process has four steps. Centaur works with the customer on dataset design, starting from the customer's raw data or just a problem statement. Experts and AI models then compete, drawn from a pool of more than 100,000 subject matter experts, and are measured and rewarded on performance. An aggregation step weights annotators, human or AI, by their track record to outperform either alone. Finished datasets are delivered through API, JSON, CSV or DICOM. Example tasks on the site include skin lesion classification, retinal OCT grading and ICD 10 code mapping. An interactive playground lets teams test a labeling strategy and compare cost against accuracy. The company grew out of research at the MIT Center for Collective Intelligence and was formerly called Centaur Labs. Its audience is machine learning teams, from startups to enterprises, that need reliable labels for medical imaging, clinical text and similar data, and the site also invites experts to join its Centaur Arena as annotators. The vendor does not publish pricing and directs prospective customers to book a demo.