CurateLM is a data curation and verification lab based in Berlin, Germany, founded in 2026. We specialise in sourcing, curating, and licensing high-quality text, biomedical, and legacy code datasets for AI and LLM training.

Our focus is on data that mainstream markets ignore: rare and underrepresented languages from Africa, South Asia, Southeast Asia, Central Asia, the Caucasus, and Europe, alongside curated biomedical text and commercially-clean legacy code. Every dataset passes through our verification lab, where we apply PII scrubbing, deduplication, license verification, and quality scoring. From an initial pool of 154 billion tokens, only 5.59 billion passed our commercial-clean standard — no NC, ND, SA, GPL, or AGPL content.

We share samples on the Mozilla Data Collective to make rare-language and niche-domain data discoverable to the open AI community, in line with Mozilla's mission to make AI training data more multilingual, multicultural, and representative. Full corpora are available under separate commercial license with controlled access.