Basalt Data
Provenance-tracked, rights-cleared corpora for regulated training.
- Sector
- Data infrastructure
- Headquarters
- Remote
- Partner since
- 2025
- Team
- 9
Overview
Basalt assembles training corpora where every document carries a verifiable chain of custody and an explicit license, so that the model trained on it can be deployed somewhere the provenance question actually gets asked.
Why this exists
Buyers in health, defense, and financial services increasingly have to answer what a model was trained on. Most training data cannot answer it. The corpora that can are assembled ad hoc by each buyer, at enormous cost, and thrown away after one project.
Basalt is building that as a product: licensed at the source, hashed at ingestion, queryable by rights terms, and delivered with a manifest that survives audit.
Why we partner
Rights clearance is slow, legally intricate, and gets cheaper per document the more of it you have already done. It is a bad venture-timeline business and a good permanent-capital one — which is why the round Basalt needed did not exist until we wrote it.
Current focus
- First licensed corpora in clinical and legal domains
- Manifest format and third-party attestation
- Design partners with a live procurement requirement