Neosync - PII Detection and Anonymization in Free-form Text

We’re an open source data anonymization and synthetic data platform that companies like Intel, Siemens, C2FO, Alasco and others use to anonymize their sensitive production data and sync it to lower-level environments.

Today, we’re launching a new product designed to detect and anonymize PII data in free-form text.

These are two main use cases:

1. Detecting and redacting PII data before sending it to an LLM for inference. If you’re building agentic systems or working with LLMs and sensitive data, you shouldn’t be sending your sensitive data to those LLMs. You can now use our API to first detect and anonymize that data and then send it to your LLM provider.
2. Detecting and redacting PII data before training. You generally want to avoid training a model on PII (especially if others will be using it). You can use our API to detect and redact free-form text in training data so that you’re not training it on PII.

For example:

The text:

{ text: "Dear Mr. John Chang, your physical therapy for your rotator cuff injury is approved for 12 sessions. Your first appointment with therapist Jake is on 8/1/2024 at 11 AM. Please bring a photo ID. We have your SSN on file as 246-80-1357. Is this correct?"}

Would be transformed to:

Anonymization result: '{"text":"Dear Mr. \u003cREDACTED\u003e, your physical therapy for your rotator cuff injury is approved for 12 sessions. Your first appointment with therapist \u003cREDACTED\u003e is on \u003cREDACTED\u003e at \u003cREDACTED\u003e. Please bring a photo ID. We have your SSN on file as \u003cREDACTED\u003e. Is this correct?"}' 

You can also customize this with custom allow/deny lists and even custom recognizers.

We’re already working with companies in Healthtech and Fintech on this and would love to open it up to more companies. If you’re interested in trying it out, shoot me a note at evis@neosync.dev, and I can get you a sandbox and free credits to trial it.