At a time when transparency and privacy must go hand in hand, a Flemish government agency faced a complex challenge: how to make thousands of documents publicly available without infringing on the privacy of those involved. The classic, manual approach of anonymizing proved no longer feasible. Therefore, an innovative project was launched, in collaboration with Möbius, to build a scalable and reliable solution using artificial intelligence (AI).
Strategic Challenge
The organization receives increasing requests for disclosure of documents, in accordance with the Public Access Act (WOB) and GDPR legislation. The existing workflow - manually covering sensitive information on thousands of pages - was slow, error-prone and labor-intensive. This led to long turnaround times and a high workload for employees. So the challenge was twofold: on the one hand, comply with legal requirements; on the other, drastically increase operational efficiency. The document issue was very diverse both in format and structure (different extensions, page sizes, scanned documents versus digital PDFs...).
Approach
The project team adopted an AI-driven approach with an architecture that combines several technologies:
- Optical Character Recognition (OCR) for text recognition.
- Deep Neural Networks for signature and photo detection
- Large Language Models (LLM) for recognition of textual sensitive personal data
- Integration with existing document management systems for smooth implementation
Stakeholders were involved from the beginning through workshops and testing. The tool was built in a modular fashion so that future extensions remain possible. A human-in-the-loop system was built in to validate uncertain outputs with human intervention. Documents were irreversibly anonymized while maintaining searchability in the documents. In this, Möbius was responsible for the project end to end: from business analysis, to engineering the AI models and bringing the AI architecture into production in the cloud, as well as integration with the client's internal systems.
Result
The results speak for themselves:
- 1000% increase in throughput
- 92.5% reduction in workload for employees
- More than 100,000 pages automatically anonymized within three months
The tool was not only developed as a proof of concept, but also fully integrated into the operation of the organization. Thanks to the AI solution, employees can now focus on more complex files while routine tasks are automated.
