TALK TO AN EXPERT
TALK TO AN EXPERT
DATA DE-IDENTIFICATION

De-identify data at scale, without scrubbing what made it valuable.

Context-aware ML detects PII, PHI, PCI and corporate confidential data across 60+ entity types and 54 languages—then redacts, pseudonymizes, tokenizes or replaces it with synthetic data. All inside your own infrastructure.

TALK TO AN EXPERT
TALK TO AN EXPERT
30-minute call with a Limina engineer. No deck, no discovery script.
For data, security and compliance teams handling regulated or sensitive data at volume.
IMPACT

0.2 - 7%

PII missed by Limina, against 13.8 - 46.5% by general-purpose tools

18/18

HIPAA Safe Harbor identifiers covered

60+

entity types across 50+ languages

0

bytes of your data sent to a third party

Why it stalls

Manual review

Accurate until volume arrives. Then it is a backlog with a headcount line attached.

Regex and rule libraries

100+ patterns, a dedicated bug channel, and no way to catch what context reveals.

General-purpose cloud APIs

Built for breadth, not for identifiers. Between 13.8% and 46.5% of PII missed—and your data has to leave your environment to get there.

Built for real-world data

Blue square with a black plus symbol partially overlaid by a tilted light blue eraser.

Detect what matters

Context-aware ML identifies PII, PHI, PCI and corporate confidential data across 60+ entity types the way a trained reviewer would. Coreference resolution links names, abbreviations, initials and pronouns to one entity, so nothing slips through and nothing gets double-counted.
Blue square with a black plus symbol partially overlaid by a tilted light blue eraser.

Choose how it is de-identified

Redact it, pseudonymize it, tokenize it reversibly, or replace it with synthetic
data that preserves the statistical shape of the dataset. Configure the method per entity type and per workflow—one rule for names, another for dates, another for account numbers.
Blue gavel with a warning icon above it symbolizing legal caution or alert.

Keep it useful, and defensible

Because "Michael," "Mike" and "M. Smith" stay linked as one token, longitudinal analysis survives de-identification. Output is deterministic: the same input always produces the same output, so the audit trail holds up. Certified to HIPAA Expert Determination standard with a re-identification risk of 0.04%.
Talk to an Expert
Talk to an Expert
CUSTOMER WIN

Boehringer Ingelheim

Minutes

Down from 2 – 7 days per inquiry

US, EU, China, Japan

Regions anonymized to one standard

Enterprise-wide

From single use case to Data Foundation layer

The knowledge existed. The compliance path didn't.

Boehringer Ingelheim's DUDE team built Smurf, a platform for answering medical
inquiries from doctors, patients, pharmacists and wholesalers. The questions
arrived as plain text and complex PDFs, dense with personal information, in
French, English, Spanish, German and more. Without proper anonymization, global
data protection compliance—US-Europe sharing above all—was impossible.

Limina cleared it.

Limina anonymizes at the source, before translation, inside Boehringer's own
architecture. What started as one use case became their enterprise Data
Foundation layer: downstream applications now consume pre-anonymized data and
never handle personal information directly. Response times fell from days to
minutes, and the US became one of the platform's largest data sources.

We couldn't have launched without proper anonymization. It was critical for compliance, especially to include US data.

Israel Gonzales Salas
Data Engineer, Boehringer Ingelheim

Your data never leaves your environment.

One container in your VPC, your on-premises infrastructure, or a fully air-gapped environment. Two Docker commands. All processing runs in memory,
nothing is written to disk, and there is no outbound call to Limina. No third-party access. No data usage rights signed away. Ever.

CONTACT US

Ready to de-identify your own data?

A 30-minute call with a Limina engineer or solutions expert. No deck. No
discovery script. We'll ask what you're working on, where you're stuck, and whether Limina is the right fit.

What you leave with

  • What accuracy to expect on your data type
  • Which de-identification method fits each entity you handle
  • A deployment path for your environment, and a realistic timeline to production

De-Identification, Anonymization, Pseudonymization

What's the difference between de-identification, anonymization and pseudonymization?

De-identification is the umbrella: removing or replacing information that could identify a person. Anonymization is irreversible—the original values are gone. Pseudonymization is reversible under controlled conditions, because a consistent token stands in for each entity. Limina does all three, configurable per entity type, so you can anonymize a name and pseudonymize a patient ID in the same pass.

Will de-identified data still be useful for analytics and AI?

That's the point. Coreference resolution keeps every reference to the same person linked to one token, so longitudinal analysis survives. Relation extraction preserves the connections between entities. Synthetic replacement maintains the statistical shape of the dataset. You get data you can train on, share and analyze—not a field full of black boxes.

Does it work on unstructured data?

Unstructured data is where identifiers actually hide—inside sentences, abbreviated, misspelled, referred to obliquely. Limina handles free text, PDFs, DOCX, images, audio and DICOM across 15+ file formats, at 70,000 words per second on GPU.

Is there an API?

Limina is a REST API deployed as a container. No SDK required — if you can make an HTTP request, you can integrate it. All configuration happens at the request level, so there's no server admin and nothing hidden behind a UI. It plugs into Airflow, Kafka, Spark or any custom ETL layer.