RedactAI.

What it does

RedactAI removes personally-identifying and protected health information from documents before they leave a trusted boundary — for legal discovery, regulator submissions, internal training data, or sharing with external partners. It is precise about what counts as identifying in your jurisdiction, not generic.

How it works

The pipeline runs a layered detection pass — pattern-based extractors for the well-known classes (national IDs, MRNs, contact details), an LLM judge for the harder cases (initials, role-implied identifiers, indirect inference), and a policy layer that decides whether each detected item is redacted, pseudonymized, or surfaced for human review. Every output file ships with a signed redaction map your compliance team can inspect.

Deployment & governance

API-first or run as a guarded batch service. Policy definitions live in version-controlled YAML; reviewer overrides are audited. Designed for regulated workflows — healthcare, finance, legal, government — where “we removed the obvious stuff” isn’t a defensible standard.