Is this the same as generating a fully synthetic dataset?

No. Full synthetic data generation creates an entirely new, statistically representative population from scratch. Contextual Replacement transforms selected values in your existing source data, preserving the record itself while replacing what identifies the people in it. They solve overlapping but different problems.

Does Contextual Replacement replace Limina's current synthetic replacement method?

No. Both options remain available, and teams can choose the method that fits each workflow. There is no forced migration and no change to existing configurations.

Which types of entitites does it support?

At launch, Contextual Replacement supports names, organizations, credit card numbers, card verification values, bank account numbers and phone numbers.

Is it a separate product or a paid add-on?

No. It's included for all Limina customers as an additional replacement option within the existing product.

Does it require new infrastructure?

No additional infrastructure is expected. The capability is included in the existing Limina container. Confirm version and configuration details in the documentation for your environment.

Does using Contextual Replacement make a dataset compliant?

Contextual Replacement can support an organization's privacy and compliance workflows, but compliance depends on the data, jurisdiction, use case and implementation. Limina can help your team determine the appropriate de-identification approach for your requirements.

How accurate is Limina compared to manual redaction and other similar products?

Limina achieves 99.5% accuracy on physician conversations, as demonstrated in deployment with Providence Health.

Does my data leave my environment?

No. Limina's container runs in your VPC or on-premise. Data never leaves your infrastructure.

What languages and data formats does Limina support?

Limina supports 52 languages and works across structured and unstructured data including text, images, audio, and documents. It handles multilingual content and code-switching.

Which compliance standards does Limina meet?

Limina provides independent expert determination reports for HIPAA, GDPR, CPRA, and other global privacy regulations.

Can I use de-identified data for AI training, analytics, and data sharing?

Yes. Limina's context-aware de-identification preserves data utility while removing sensitive information, making it suitable for AI training, analytics, and compliant data sharing.

CONTEXTUAL REPLACEMENT

Protect the Identity.
Keep the Context.

Q: What is Contextual Replacement?

Contextual Replacement is a Limina capability that substitutes sensitive information with realistic synthetic data selected according to the surrounding context. Rather than swapping a detected value for a random or generic stand-in, it produces a replacement designed to fit the record it appears in, helping protected data retain relevant characteristics for approved research, analysis and data-sharing workflows.

Replace sensitive identifiers with contextually accurate synthetic values, so your data stays useful for research, collaboration and AI training while it stays inside your own environment.

REQUEST A DEMO

Original · fictional record

Patient Clara Montgomery was admitted to Lakeside General Hospital in Seattle. Referred by Dr. Howard Klein.

With Contextual Replacement

Patient Zhuan Yinihong was admitted to National General Hospital in Glasgow. Referred by Dr Connor David.

Trusted for

HOW IT WORKS

From protected to useful

One hybrid request replaces every detected identifier, three steps from sensitive source data to output your teams and models can use.

Blue magnifying glass icon with a light blue translucent square behind the lens.

Detect what matters

Limina's context-aware NER identifies sensitive entities across unstructured text and files, understanding language the way a human reader would.

Stacked blue translucent squares with rounded corners in a layered diamond shape.

Replace with context

Model-based generation creates contextually accurate replacements for names, organizations, locations and origin. Deterministic rules handle dates, phone numbers and financial identifiers. Everything else falls back automatically to Limina's existing synthetic system.

Blue shield icon with a white checkmark symbolizing security or compliance verification.

Use with confidence

The output stays coherent: readable to reviewers, consistent for downstream systems and ready for research collaboration, partner sharing and AI training.

Context, preserved by design

Higher-quality synthetic replacements for the workflows where transformed data still has to make sense.

Simple blue outline of eyeglasses with round lenses and curved temples.

Context stays intact

Replacements fit their surroundings. Organizations stay within their category, physicians remain physicians and records keep reading like records.

Related names stay related

Family members can keep a consistent surname after replacement, so the relationships your research depends on survive de-identification.

Blue clock icon with hour and minute hands inside a rounded rectangle border.

Valid structure, every time

Deterministic rules produce dates, phone numbers, addresses, bank accounts, credit cards and routing numbers with realistic, valid structure.

Simplified atomic model with nucleus and electron orbitals in blue outline.

Automatic fallback, complete coverage

Supported entities use the enhanced hybrid system. Every other detected entity falls back automatically to Limina's existing generator in the same request.

Icon of a certificate with lines of text and a ribbon seal on a rounded square background.

Your data never leaves your environment

Contextual Replacement runs in the GPU Synthetic Enhanced container in your VPC, on-prem environment or any major cloud.

SEE IT IN CONTEXT

REQUEST A DEMO

See the difference context makes

Every identifier replaced

Names, organizations, locations, dates, phone numbers and financial identifiers, all synthetic.

Relationships preserved

Related names stay related and organizations stay within their category.

Still reads like a record

Coherent to reviewers, consistent for systems, useful for AI training.

Original · fictional record

Patient Clara Montgomery was admitted to Lakeside General Hospital in Seattle on July 19, 2025. Contact number on file is (206) 555‑0192. Referred by Dr. Howard Klein.

With Contextual Replacement

Patient Zhuan Yinihong was admitted to National General Hospital in Glasgow on July 18, 2016. Contact number on file is (781) 626‑8889. Referred by Dr Connor David.

GET STARTED

Ready to see it on your data?

Bring a research, collaboration or AI-training workflow and we'll show you what Contextual Replacement does with it. Included for all customers on release 4.5.

Blue circular badge with AICPA SOC and URL aicpa.org/soc4so on it.

ISO 27001 Certified badge for Information Security Management.

Shield icon with a blue hexagonal network symbol and the word VERIFIED below it.

Frequently Asked Questions

What is Contextual Replacement?

A hybrid Limina capability in release 4.5 that uses contextual model generation and deterministic rules to create higher-quality synthetic replacements for supported identifiers. Replacements are designed to fit surrounding context and preserve selected relationships. It is included for all customers on release 4.5 and is opt-in; the existing replacement system remains the default.

Which entity types are supported at launch?

Model-based generation covers LOCATION, LOCATION_STATE, NAME, NAME_FAMILY, NAME_GIVEN, NAME_MEDICAL_PROFESSIONAL, ORGANIZATION and ORIGIN. Rule-based generation covers DATE, LOCATION_ADDRESS, LOCATION_ADDRESS_STREET, LOCATION_CITY, LOCATION_COORDINATE, LOCATION_COUNTRY, LOCATION_ZIP, PHONE_NUMBER, BANK_ACCOUNT, CREDIT_CARD, CREDIT_CARD_EXPIRATION, CVV and ROUTING_NUMBER. All other detected entity types automatically fall back to Limina's existing synthetic generation system within the same request.

Is this the same as synthetic data generation?

No. Contextual Replacement transforms identifiers in existing source data. It does not generate an entirely new, statistically representative population from scratch.

What infrastructure does it require?

Release 4.5's GPU Synthetic Enhanced image, which includes existing GPU Synthetic functionality plus the new capability. It deploys like Limina's other GPU containers but requires a single GPU with 24 GB+ VRAM and a longer startup allowance. Follow the documented VM/GPU and Kubernetes guidance. Available on supported process routes, including process/text and process/files; not available on analyze/text or in CPU images at launch.

How do I enable it, and how do the accuracy modes differ?

Set synthetic_entity_accuracy to standard_high_multilingual (the best quality and speed balance, a good default for larger volumes) or high_multilingual (the best possible output quality via model reasoning, considerably slower). These synthetic accuracy modes are independent of NER detection accuracy.

Can the output be shared freely?

Not automatically. Sharing decisions should reflect the full dataset, configuration, intended use, applicable requirements and your organization's privacy review.

Protect the Identity. Keep the Context.