How we source, de-identify and license data.
What happens before a dataset reaches an AI team. Written for data and legal teams.
Sourcing
- We source across industries and data types, and we keep adding companies.
- We work with established businesses that have years of continuous records across several connected systems.
- Before anything moves, the company confirms in writing that it owns the records and can license them.
- Each company can leave out systems, people, date ranges or topics.
What we don't take
- Patient and medical records
- Biometric data
- HR and payroll systems
Linking
- Records are linked across systems, and the links survive de-identification.
De-identification
- Consistent pseudonyms
- People and companies get realistic stand-ins that stay the same across systems and years.
- Typed tokens
- Account, phone and ID numbers become labeled tokens.
- Handling
- AI teams receive de-identified data, never the original records.
What ships with a dataset
- Messages and records in JSONL or Parquet.
- A schema, record and token counts per system, date coverage, and documentation of the de-identification method.
License terms
License- Training and evaluation
- No attempts to re-identify people or companies
Terms are set per dataset, in writing.
Source company
Layer Gray
AI team
Questions from your data or legal team?
Write to contact@layergray.com.