OVERVIEW
Handling sensitive judicial documents safely is a complex task. It is not enough to remove names or match keywords. In many cases, the context determines whether information should be masked before a document can be shared.
To explore this challenge, GESHDO built an application designed to process preliminary investigation documents and identify sensitive information before export. The solution was shaped around the types of requirements found in public sector environments, where privacy, legal handling, and trust are essential.
GESHDO owned the full project end to end, from project management and solution design to implementation. The application was built using the BART model to interpret text in context and identify information that should potentially be masked before release.
Instead of relying only on static matching, the solution was designed to better understand how names and sensitive details were used within the surrounding content — and then present masking suggestions to the user before PDF export.
OUR APPROACH
We built a practical AI workflow that supports human review rather than replacing it.
Context-Aware AI
The BART model was used to interpret text in context, going beyond keyword matching to understand how sensitive information was used within the surrounding document content.
Human in the Loop
The product presented masking suggestions to the user rather than automatically redacting. This kept humans in control of the final decision — critical in high-trust environments.
Privacy-First Design
The solution was built with privacy-aware principles throughout, designed for use cases where sensitive information must be handled carefully and responsibly.
PDF Export Workflow
Once the user reviewed and confirmed the masking suggestions, the document was exported to PDF — creating a clean, auditable workflow from input to output.
OUR RESULTS
A strong example of applied AI in a complex, regulated, and trust-sensitive domain.
Working Application
The result was a working AI-based application for identifying and masking sensitive information in judicial documents — practical and usable from day one.
Proven Concept
The solution proved that context-aware masking could be implemented in a practical and usable way, moving beyond simple keyword replacement.
AI in Regulated Environments
The project demonstrated GESHDO's ability to build AI applications for complex, regulated, and trust-sensitive scenarios — where both technical quality and ethical design matter.
Idea to Product
This case is a strong example of applied AI, product thinking, and the ability to move from idea to working solution in challenging domains.