DatologyAI

Datology is a data curation platform for teams that train AI models. The vendor describes it as a data refinery that turns a customer's data into high quality training data for mid training on open models or pre training their own foundation models. It runs on the customer's data and inside their private cloud, and the vendor states that it does not sell data or source tokens. The product is Datology Curation Studio, which runs a four stage pipeline. Clean removes malformed, empty, short and benchmark contaminated documents at petabyte scale. Calibrate tailors data by quality, task and taxonomy. Create generates synthetic data to improve task relevance and diversity. Compose mixes data sources into staged training datasets. Inputs can be proprietary enterprise data, public web data and licensed data from partners. It is aimed at AI model builders and enterprise teams. The vendor cites a legal domain adaptation case where mid training on a proprietary corpus lifted Legal Bench results, and a frontier open weights mixture of experts model built on 17 trillion tokens curated by Datology. The company also publishes research on data curation. No pricing is published, and access starts with booking a demo.