While large parts of the private sector are beginning to use cloud applications like ChatGPT in knowledge management, public authorities face significantly higher security requirements. To ensure data sovereignty and cyber resilience, dependencies on external providers such as Microsoft, OpenAI, and others must be minimized. A governmental system must be operated on-premises to ensure maximum sovereignty.  

SecuredGPT focuses on the development of an internal knowledge management system based on large language models, specifically tailored for use in the public sector. The system is designed to assist law enforcement in answering complex queries in real-time by accessing specific internal knowledge bases, such as the ‘Kriminalistischer Leitfaden’ (KLF).  

The project includes several phases: first, an analysis of suitable language models with a focus on German language comprehension is conducted. Based on this, the system is integrated with internal documents and evaluated in a pilot deployment with a selected group of test users. Through iterative optimization based on user feedback, the chatbot will be technically and functionally adapted to meet practical operational requirements.  

Key objectives of the project include gathering insights for a potential production rollout and ensuring on-premises compatibility. The primary focus areas are data security and transparency regarding data sources. The pilot deployment provides both qualitative evaluations and insights into runtime performance in an on-premises environment.

Project Lead
Cortecs GmbH
Schottengasse 10
A-1070 Wien

Partners
Bundesministerium für Inneres

Contact
Markus Tretzmüller, MSc.
+43 680 3230395
markus.tretzmueller(at)cortecs.ai 
https://cortecs.ai