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, proprietary curation platform deployed into customer VPC; no source-available product. Founded 2023 by Ari Morcos (ex-FAIR/DeepMind) and Matthew Leavitt (ex-MosaicML). ~$57.5M raised (seed + $46M Series A); angels include Jeff Dean and Yann LeCun. No named frontier-lab customers disclosed. Confirmed 2026-07-21.

Openness

1 high confidence
1.0
license
Proprietary
source
closed

Fully proprietary automated-curation engine; the closed analogue to datatrove/NeMo Curator, science-of-data-curation niche.

Adoption

2 medium confidence
2.0

~$57.5M raised with high-profile backers, but early-stage with no named frontier-lab deployments; traction is credibility/funding rather than proven scale. Directional.

Capability

3 medium confidence
3.0

Strong specialized curation engine, narrow scope and early adoption.

Unchanged since 2026-07-30 (last edited, not re-checked)