DatologyAI

DatologyAI builds an automated data-curation platform for teams training foundation models or adapting open models to their domains. Deployed in customers’ private clouds, its pipeline cleans data, selects material by quality and task, generates synthetic examples, and mixes sources into staged training datasets. It works with customers’ enterprise, public-web, and licensed data rather than selling or sourcing data itself. The aim is to improve model performance while reducing the compute needed for training.

The company’s co-founders are CEO Ari Morcos, CTO Bogdan Gaza, and Matthew Leavitt. Its research includes BeyondWeb, the synthetic pretraining data component of its curation pipeline. BeyondWeb rephrases existing documents into varied formats and styles, including question–answer pairs, to improve information density and relevance. Its experiments found that diversity in generation strategies helps sustain learning during extended training, while simply matching conversational styles delivers more limited gains.

Customer work illustrates the platform’s operating scale: in 2026, DatologyAI reported preparing a 20-trillion-token pretraining dataset for Arcee AI, including more than eight trillion synthetic tokens. The company raised a $46 million Series A led by Felicis Ventures in 2024, bringing total capital raised to more than $57.5 million.