Sudhish Swain invited you to de-identify your documents

Founder, Petrarch · sudhish@petrarch.co · sent to Marcus Thackeray at Northwind Resources

Before you send the claims backfile over, run it through this. It strips names, addresses, account numbers and health details, and keeps the file readable. Run it on your own machine if you would rather we never see the originals - both routes below are the same engine.

What Petrarch Lindo does. It reads a document and finds the people, addresses, emails, phone numbers, account and record numbers, and organisations in it. Each one is either replaced by a consistent token (C-7o08r, the same token everywhere that value appears, so your data still joins) or by a realistic stand-in (Northwind Resources, LLC, so the file still reads like a file). You decide which, per type, and you can overrule any single decision.

Two ways to run it. Both are the same engine.

Run it here, in your browser

Nothing to install. Works on any machine, including a locked-down laptop.
Handles a whole bucket, not just one file, and shows the cost before it runs.
We hold the document while it processes. Encrypted in transit and at rest in our AWS account, deleted 7 days after upload, and Petrarch holds the encryption key. Staff access is audited, not impossible.

Download the app and run it locally

Nothing leaves your machine. The models run offline. You can pull the network cable and it still works. We never see the document, so there is no key for us to hold.
Same detectors, same settings file, same output as the cloud run.
It needs hardware. 16 GB of RAM, Apple silicon or a 2019-or-later x86 machine, and a 1.8 GB download. A large corpus takes hours where the cloud takes minutes, and you run it, so you own the babysitting.

Download for macOS2.4.1 · Apple silicon · 1.8 GB

Not sure which? Compare where it runs - our cloud, your own VPC over PrivateLink, or on-prem - side by side, including who holds the keys and what we can see in each.