Welcome to the TekAI demo portal.

The heart of the demo: upload a PDF or text document or paste text directly — detection finds names, email addresses, phone numbers, IBANs and AHV numbers. The before/after view highlights every match in color with its confidence score; choose between redaction, placeholders and pseudonyms, and download the masked file. For PDFs the original text is truly removed, not just covered.

How AI agents work safely with your data: your prompt is masked before it leaves the building, the LLM answers based on placeholders, and only in the final step is the response unmasked again. All four steps are shown individually — including the “what the LLM sees” view proving that no sensitive data is transmitted.

How reliable is the detection? This page shows precision, recall and F1 score for each entity type individually, plus pipeline latency for text and PDF processing — measured on our Swiss evaluation dataset with documents in German, French, Italian and English.

Which data deserves how much protection? From specially protected personal data (health, religion, biometrics) via identifiers such as the AHV number and IBAN to general personal data like names and email addresses — classified under the nDSG and GDPR, and what that means for masking.

The legal framework: what the Swiss nDSG, the GDPR and the EU AI Act mean for using LLMs with personal data — and why masking before transmission is a key building block. Plus the privacy principles of this demo: in-memory processing only, no storage, no content in logs.