The Unified Intelligent Assistant is an AI powered conversational tool developed by the Ministry of Digital Transformation that is integrated into existing websites to help users quickly find reliable information from government portals. It reduces the burden on traditional support channels by automating answers to recurring questions and provides 24/7 access to clear, user friendly information.
Initiative overview
The Unified Intelligent Assistant is a pilot AI solution launched by the Ministry of Digital Transformation in 2025 to modernise public sector support services. Traditional channels such as phone lines and email are frequently overloaded, leading to delays and lower user satisfaction. The Assistant addresses this challenge by enabling fast, intuitive access to information through natural language conversation, consolidating content that is currently dispersed across multiple government portals.
The Assistant supports the strategic goals of the “Digital Public Services 2030” agenda and is part of the broader “Development of New Dynamic e Services (DES)” programme. Its purpose is to provide a unified, reliable, and secure digital helper that improves information accessibility, reduces support workloads, and ensures around the clock availability. The first data source included in the pilot phase is the SI TRUST portal, with planned expansion to additional government portals.
The assistant operates as a web based conversational interface that is integrated into existing websites, powered by a large language model, returning answers strictly based on predefined public sector content. Responses always include references and links to original sources. The Assistant functions exclusively in Slovene, using formal address, maintains conversational context within a session, and deletes session data after 30 minutes of inactivity or upon closing. It can recognise questions posed in different languages. It does not require login, does not perform transactions (e.g. filing applications), and serves solely informational purposes.
A multi-layer anonymisation mechanism ensures compliance with data protection requirements. Since users might inadvertently share personal data despite warnings, all inputs undergo structured and unstructured personal data detection using regular expressions and named entity recognition (spaCy), followed by automatic removal before being sent to the LLM.
Results, outcomes and impacts
During the testing phase, the Unified Digital Assistant proved highly effective in providing a wide range of accurate information within the supported domains. It is expected to significantly reduce user calls and emails to the support centre (EKC) and decrease the number of user errors when submitting documents for public services. As the initiative is still in its pilot phase, measurable indicators and formal targets have not yet been defined.
Other relevant details
Challenges and lessons learned: Key challenges included ensuring compliance with data protection rules, preventing accidental submission of personal data, and maintaining answer accuracy across different content domains. These issues were mitigated through a multi-layer anonymisation process and a knowledge base limited to verified public content. A significant lesson learned was the difficulty to reliably recognise different types of personal data in a system that must interpret diverse linguistic patterns; several Slovenian words were mistakenly identified as personal names from other languages, resulting in false positives and unnecessary redaction. This highlighted the need for continuous tuning of anonymisation models and contextual checks. Managing diverse and evolving content across multiple portals also requires coordinated governance, while lessons learned emphasise the importance of clear user guidance, ongoing quality monitoring, and close collaboration with content owners to ensure long term scalability.
Additional insight: The Unified Intelligent Assistant represents an important step toward a user centric digital public administration and demonstrates how artificial intelligence can be responsibly integrated into government services. Beyond its core functionality, the initiative serves as a practical testbed for developing standards, governance models, and technical frameworks for future conversational AI tools in Slovenia’s public sector. The Ministry plans to use insights from the pilot to define long term operational models, including content governance procedures, quality assurance workflows, and cooperation mechanisms with other public institutions. This will support the eventual scaling of the Assistant into a shared government solution, enabling consistent user experience across multiple portals and service domains. A key value of the initiative lies in its robust approach to data protection and safety. The multi layer anonymisation process—combining rule based checks, named entity recognition, and context analysis—ensures that personal data is removed before any interaction with the language model. This design choice not only mitigates privacy risks but also sets an example of best practice for AI deployment in sensitive public sector environments. The project also emphasises transparency by informing users about the purpose and limits of the tool and by ensuring that all responses are grounded exclusively in official, publicly available content.
