People
Names and personal mentions in text.
⟨PERSON⟩
// Persona · Built into Mindverse Studio
More possibilities for AI. More protection for your data.
Use AI for sensitive tasks, too. Our integrated pseudonymisation uses a fine-tuned model to detect personal data and replace it before your text reaches the downstream AI model.
- Technology
- Fine-tuned model
- Detection
- 9 data categories
- Processing
- In Germany
Persona in actionFine-tuned model
01 / Your original text
Please reply to Anna Weber’s request. Her customer ID is KD-20481. Contact: anna.weber@example.com.
Persona detects & replaces
02 / What the LLM sees
Please reply to ⟨PERSON_1⟩’s request. Her customer ID is ⟨NUMBER_1⟩. Contact: ⟨EMAIL_1⟩.
Interactive example · Fictional data
Built into Studio
Context-aware detection
Consistent pseudonyms
Re-identification
// How pseudonymisation works
Keep the meaning. Replace the identity.
Persona adds a protection step to your AI workflow, from detecting sensitive details to delivering a usable response.
01
Detect sensitive data
The specialised model analyses text in context to identify personal details such as names, addresses and account numbers.
02
Work with pseudonyms
Detected details are replaced with consistent placeholders. The downstream LLM processes the pseudonymised text while retaining useful relationships.
03
Restore the response
A separate mapping lets you replace pseudonyms in the response with the original details, turning the result into a usable draft.
// A dedicated model for data protection
Fine-tuned for sensitive data. Trained to understand context.
Personal information often sits inside ordinary prose. We use a specialised, fine-tuned language model to recognise details in context, including in German text.
Detection in context
The model considers how a detail is used rather than relying solely on fixed character patterns.
Relationships stay readable
Consistent pseudonyms keep people and their statements connected, so the AI can still work with the content.
Analysis on our own hardware
The Persona model runs on dedicated servers in Germany. No external AI service is called for detection.
Context beyond individual characters
Please inform Anna Weber about the draft contract. Ms Weber is our contact person.
Please inform ⟨PERSON_1⟩ about the draft contract. ⟨PERSON_1⟩ is our contact person.
Illustrative example: two mentions are linked to the same person.
// What Persona detects
Nine categories of sensitive information.
From personal contact details to credentials, one specialised model covers different types of sensitive information.
Email addresses
Personal and business email addresses.
⟨EMAIL⟩
Phone numbers
Mobile, landline and international numbers.
⟨PHONE⟩
Addresses
Streets, house numbers and home addresses.
⟨ADDRESS⟩
URLs & profiles
Personal links and social media profiles.
⟨URL⟩
Dates
Birth dates and identifying appointments.
⟨DATE⟩
Account & customer IDs
IBANs, insurance and contract numbers.
⟨NUMBER⟩
Credentials
Passwords, API keys and tokens.
⟨SECRET⟩
Other sensitive details
Such as ID numbers, number plates and health data.
⟨PII⟩
// For everyday work
Sensitive tasks are part of business. Protection should be, too.
01
Customer service
Summarise requests and draft replies while detected contact details are replaced with pseudonyms.
AI for customer service 02People & HR
Work with applications and HR texts without sending detected names and personal details to the downstream LLM.
AI for HR teams 03Legal & contracts
Structure draft contracts and review their contents while detected client details and contract identifiers are pseudonymised.
AI for law firms// Part of your security approach
Data protection starts before the prompt.
Pseudonymisation complements the security features of Mindverse Studio. Persona processes texts in Germany, and its stateless API does not persist text contents or pseudonym mappings.
Explore Mindverse security// For your own applications, too
The technology behind the feature: Persona.
Want to add pseudonymisation to your own products or AI workflows? Persona is also available as a standalone API with detection, mapping, re-identification and a sensitivity score. Enterprise projects can deploy it in their own infrastructure.
// FAQ
Pseudonymisation, explained
Is pseudonymisation built into Mindverse Studio?
Yes. Mindverse Studio uses Persona technology for pseudonymisation with a fine-tuned model. Book a demo to explore the feature for your use case.
How does pseudonymisation differ from anonymisation?
Pseudonymisation replaces identifying details. A separate mapping makes it possible to restore the originals. Pseudonymised data remains personal data. Anonymisation, by contrast, aims to rule out identification permanently.
Why does Mindverse use a fine-tuned model?
A specialised language model can detect sensitive details in context. This helps with information that cannot be reliably described by fixed patterns alone, such as names within prose. The Persona model also supports German text.
Can Persona guarantee detection of every sensitive detail?
Automated detection can make mistakes or miss details. Pseudonymisation is therefore one part of your data protection approach. Your use case, model selection and appropriate review processes remain essential.
Where does Persona process texts?
The Persona API and detection model run on dedicated Hetzner hardware in Falkenstein, Germany. No external AI service is called for detection. This is separate from the downstream LLM that processes the pseudonymised text.
Does Persona store texts or use them for training?
The Persona API processes texts statelessly and does not persist their contents. The mapping is returned in the API response. Customer texts are not used for model training. These statements apply to the Persona API; storage in your Studio workspace is separate.
// Mindverse Studio + Persona
Make data protection part of everyday AI.
Explore integrated pseudonymisation and talk to us about how it can support your workflows.
GDPR-COMPLIANT · SERVERS IN GERMANY · SOC 2-ORIENTED PROCESSES