From 6 Months of Internal Struggle to a Working PoC in 3 Weeks

A Dutch notarial SaaS platform wanted to integrate AI-powered document translation and plain-language explanations into their product. The goal: make dense legal documents accessible to clients who aren’t legal professionals.

But there was a fundamental blocker. Notarial documents contain highly sensitive personal data — names, addresses, financial details, BSN numbers. Under GDPR and the emerging EU AI Act, sending this data to external LLM providers was a non-starter. The platform’s own team had spent 6 months trying to build an internal pseudonymization solution and hadn’t reached production quality.

KlusAI designed and built a pseudonymization pipeline that processes documents through three stages:

1. Pre-processing — Documents are analyzed and all personally identifiable information (PII) is detected and replaced with consistent pseudonyms. Names become “Person A” and “Person B”, addresses become generic placeholders, financial details are masked — all while preserving the document’s logical structure.

2. AI Processing — The pseudonymized document is sent to the LLM for translation or simplification. Because no real PII exists in the text, this step is fully GDPR-compliant.

3. Post-processing — The AI output is de-pseudonymized: original PII entities are re-inserted into their correct positions, producing a final document that reads naturally with all real names and details intact.

A complete audit trail tracks every pseudonymization and de-pseudonymization operation, satisfying both GDPR accountability requirements and EU AI Act transparency obligations.