AI Potluck
Model components / Dataset Processing Tools

DatologyAI

DatologyAI

Automated training-data curation platform: deduplication, model-based quality filtering, embedding-based selection and pruning, data ordering, and synthetic-data steps for text and multimodal pretraining corpora.

DatologyAI, a curation platform deployed into the customer's own cloud. Founded in 2023 by Ari Morcos and Matthew Leavitt. No named frontier-lab customers are disclosed, which is why the adoption axis has nothing to band. Verified 2026-08-13 via datologyai.com.

Openness

1 high confidence
1.0
license
Proprietary
source
closed

A fully proprietary automated-curation engine, VPC-deployed, with no source available and no open core - the closed analogue to datatrove and NeMo Curator in the science-of-data-curation niche.

Adoption

2 medium confidence
2.0

About $57.5M raised with high-profile backers - the Series A announcement figure, and the only funding number this record has - but the company is early-stage with no named frontier-lab deployments. That makes the traction credibility rather than proven scale, so the level is directional.

Capability

3 medium confidence
3.0

Strong specialized curation engine, narrow scope and early adoption.

  • https://www.datologyai.com recorded 2026-08-13

    the data refinery is described as clean, curate, create and compose stages - dedup and formatting removal, quality and taxonomy selection, synthetic generation and mixing

Verified 2026-08-13