Limina Data De-identification

Limina Data De-identification is a PII detection, redaction, and replacement tool for unstructured and semi-structured text. It is deployed as a containerized solution (single Docker container) on-premises, accessible via a REST API, ensuring data never leaves the user's infrastructure.

The product uses transformer-based ML models to identify over 50 entity types of personally identifiable information (PII) across 52 languages. Detection is context-aware, covering both direct and quasi-identifiers without relying on regex patterns.

Key capabilities include:

• PII Identification: Detects 50+ entity types (e.g., names, addresses, SSNs, credit card numbers, medical data) across 52 languages using contextual ML models.

• Text De-identification: Replaces detected PII with unique markers (e.g., NAME_1, CVV_3) or mask characters, processing up to 70,000 words per second.

• Synthetic PII Generation: Replaces real PII with contextually appropriate fake data to reduce re-identification risk, without exposing original data to the synthetic generator.

• Tokenization & Pseudonymization: Replaces PII with encrypted, reversible tokens, preserving data utility while protecting sensitive information.

• Configurable Output: Supports enabling/disabling specific entity types, block/allow lists, and custom entity types via few-shot learning.

The product is designed to support compliance with privacy regulations including GDPR, HIPAA, CPRA, and LGPD. It can be integrated as a filter into existing data workflows or databases.