Nothing personal survives the pipeline.
Three steps from patient document to a record we can keep.
- 01
Patient document arrives
In an isolated environment, names, DOB and IDs still on it
- 02
Personal details removed
All 18 identifier types stripped
HIPAA Safe Harbor identifiers
- 03
Only a clean copy is stored
We never store unredacted patient data
Encrypted delivery
Every record takes the same four steps.
SFT data, evals and RL rewards all go through it before they ship.
- 01
Source
De-identified records from clinics, labs and partners, with consent and provenance on file.
- 02
Two blind reads
Two clinicians annotate each item without seeing the other’s work.
- 03
Adjudicate
A senior reader settles every field where the two reads disagree.
- 04
Deliver
Training sets, evals and rewards come with a record of who read what.
What your legal team will ask, answered.
Provenance
- Every dataset traced to its source and region
- A complete history for every file, from intake to delivery
- Each delivery signed off by name, with a full manifest
De-identification
- Names, IDs and dates removed before anyone reads a record
- Text burned into scans erased from the image itself
- Every file checked by a person before it can leave
Regulation
- US: HIPAA Safe Harbor de-identification
- EU: GDPR-aligned processing
- ISO 27001 and SOC 2 controls
Security and access
- Experts only ever see de-identified data
- Locked review screens, every action logged
- Encrypted, and stored in India
Documents we share under NDA
- Data processing agreement template
- De-identification method note
- Sample datasheet
- Partner attestation letters