On-device demo
De-identify clinical text, on your device
Paste a clinical note and watch protected health information get removed, entirely in your browser. Nothing is uploaded, nothing leaves your machine.
Runs entirely in your browser. Nothing is uploaded.Optional, downloads a one-time model and runs locally.
Clinical text
Use synthetic data only. This is a demonstration.
De-identified output
DISCHARGE SUMMARY Patient: [NAME] MRN: [MRN] DOB: [DATE] SSN: [SSN] Address: [ADDRESS], Springfield Phone: [PHONE] Email: [EMAIL] Seen by: [NAME] on [DATE]. Mr. [NAME] presented with chronic myeloid leukemia and Type 2 diabetes. Started on imatinib 400mg daily. Follow-up scheduled for [DATE]. Insurance ID: [ID].
12 identifiers removed
NAME 3MRN 1DATE 3SSN 1ADDRESS 1PHONE 1EMAIL 1ID 1
How this demo works
Processing stays on the device
A pattern layer runs instantly and locally on the text you paste, catching structured identifiers like SSNs, dates, and phone numbers with no API call and no server round-trip. You can optionally enable an in-browser AI model (a one-time download) to catch names it misses by context. Either way, sensitive clinical text never leaves the device.
In a real engagement
What we implement for clients
- Full clinical NER models, not just pattern matching, running on your own infrastructure (on-prem, air-gapped, or mobile).
- Coverage of the 18 HIPAA Safe Harbor identifiers with configurable confidence thresholds.
- Format-preserving synthetic replacement, date shifting, and consistent pseudonymisation.
- Governance: audit reports, redaction previews, and human review on low-confidence spans.
- Validation against your own documents before anything goes live.