Materna and deepset: Sovereign AI for administration

‍

‍

(Translated from the Original Article)

Key Takeaways

• Materna and deepset bring AI applications for administration into secure and sustainable production operation.
• The open framework Haystack keeps authorities independent; transparent sources make results verifiable.
• In the pilot project of the Borken district, the time required for the entire application process decreased by around 20 percent.
• Reusable AI building blocks can be transferred to other administrative processes; Materna handles integration and operation.

Two partners, one shared perspective on sovereign AI

How can successful AI applications be transferred to other administrative processes without having to start from scratch each time?  

‍

Berlin-based AI company deepset provides Haystack, an open framework that connects language models, knowledge sources, and other components for AI applications. Haystack is the established AI orchestration standard in the German AI stack and enables, among other things, the development of AI agents that can handle complex tasks in several coordinated steps. The Haystack Enterprise Platform, built upon this framework, adds functionalities for development, governance, and operational management.

‍

Materna brings its many years of experience in the digitalization of public administration to the table. The company understands the technical requirements, established IT landscapes, and specific conditions of public authorities. Materna provides consulting services, develops suitable solution architectures and its own AI solutions, and integrates AI applications into existing specialized procedures and processes. Furthermore, it possesses expertise in secure and sustainable operation. This allows reusable AI solutions to be adapted to the specific requirements of individual public authorities and transferred to other administrative processes.

Open technology for independent management

Haystack enables the combination of different language models, databases, and other components. This allows public authorities to select suitable technologies, connect existing systems, and adapt their applications to changing requirements. As a result, they remain agile in the long term and can decide for themselves which systems they use and how they process their data. The open technology approach supports various operating models, such as sovereign cloud environments, on-premises data centers, or highly secure infrastructures.

‍

Furthermore, Haystack allows complex processes to be broken down into individual, controllable processing steps. AI agents can analyze documents, compare information, access knowledge bases, and prepare results for further processing. Evaluation and monitoring functions, as well as defined security mechanisms, support quality assurance. Transparent sources make the results verifiable.

‍

This traceability is essential for administrative procedures. When processing applications or applying legal regulations, the responsible employees must be able to recognize the information on which a result was based.

Borken district: How agentic AI relieves the burden on caseworkers

The potential of this technology is demonstrated by a pilot project in the Borken district with deepset. Within the framework of the Federal Ministry for Digitalization and State Modernization's Agentic AI Hub, deepset tested a solution for AI-supported processing of applications for long-term care assistance.  

‍

Processing such applications is complex. Individual cases comprise more than 100 pages with numerous supporting documents. Employees must review documents, compile information, and consider legal requirements. A significant portion of their working time is spent on preparatory tasks.

‍

The AI-powered system checks applications for completeness, analyzes submitted documents, prepares financial data for asset verification, and generates suggestions for further processing. Professional responsibility and the final decision regarding the application remain with the responsible staff.

‍

The results of the three-month pilot phase: For selected work steps, the time required decreased by 35 percent. This corresponds to a saving of approximately 20 percent across the entire application process.

From individual applications to reusable AI building blocks

The approach tested in the Borken district also offers potential for other administrative procedures. Document review, data comparison, and the development of specialized knowledge are tasks common to numerous public authorities. Once developed, AI components can serve as reusable building blocks. For new applications, they are linked to the relevant legal frameworks, knowledge sources, and specialized procedures.

‍

The common technological foundation is retained, while specific specialist functions are added. Materna handles the integration into the respective administrative and IT landscapes, taking into account the requirements for security, governance, and operations.

KIAN: Materna makes nursing knowledge accessible with deepset technology

KIAN demonstrates how the interplay between Materna and deepset works in a concrete application. The digital advisor for care services is part of Materna Sinfonie , a platform for empowered digital humans.

‍

Sinfonie combines AI orchestration and knowledge discovery based on Haystack with an intuitive, dialogue-oriented user interface. While specialist information is processed and AI agents are coordinated in the background, people can communicate with the application using interactive avatars in natural language.

‍

KIAN draws on a curated knowledge base on the topic of care. Using Retrieval-Augmented Generation (RAG), relevant specialist information is specifically retrieved and incorporated into the answers. KIAN provides information on potential entitlements, care benefits, and application requirements. Furthermore, the digital advisor assists in checking the completeness of applications and points out outdated information. Source citations and the respective information status ensure that the information is verifiable.

Sovereign AI as the basis for administrative modernization

Open technologies and reusable components make it easier to transfer solutions to other authorities and specialized procedures. At the same time, the applications can be adapted to different professional, organizational, and technical requirements.

‍

“Sovereign AI will only become truly meaningful when organizations can deploy it in production on their own terms. Together with Materna, we are bringing open, European AI technology to precisely those environments where control, trust, and reliability are paramount,” says Milos Rusic, CEO and co-founder of deepset, regarding the partnership.

‍