PSA: document extraction and redaction of sensitive data — locally
Jointly awarded the Constantinus Award 2026
Customer: PSA
The task
At PSA, customer documents arrive continuously and need to be read, structured, and processed further — and they regularly contain sensitive personal data. Both needed to happen automatically, without documents ever leaving the company’s own network.
The solution
On the rzfz.ai Stack — the integrated open-source stack for local AI infrastructure — a local pipeline processes incoming documents: Docling reads and interprets the content (including OCR), Microsoft Presidio detects personal data and redacts it automatically, and Dify orchestrates the workflow. All models run on the company’s own hardware; no document and no prompt ever leaves the infrastructure.
The result
Incoming documents are automatically extracted and cleared of sensitive data before people or downstream systems work with them. The project won the 2026 Constantinus Award in the category “Standardsoftware und Cloud Services” — Austria’s most prestigious consulting and IT award.