GUARDIAN – When AI Assists Quietly in the Background
Demographic change and a growing shortage of care professionals are putting companies in the healthcare sector under increasing pressure to innovate. The research project GUARDIAN demonstrates how modern AI methods can help develop scalable, privacy-compliant assistive systems for home-based care and which technologies make the decisive difference.
The project develops a networked indoor sensor system that detects activities contactlessly and links them to an intelligent voice assistant. The system responds to situations as they arise: it reminds, alerts, activates and continuously learns in the process. Technologically, GUARDIAN builds on multimodal sensor data fusion, Large Language Models (LLMs) and context-adaptive interaction logic, developed according to the principle of Privacy-by-Design.
DFKI, with its Lübeck-based research department AI for Assistive Health Technologies (AGT), is responsible for the AI architecture of the project. The focus lies on AI-driven Activity Recognition (Human Activity Recognition), the integration and control of large language models, and the development of robust, modular interaction pipelines. The architectures developed are deliberately modular and domain-independent, making them directly transferable to other industries and application areas. Project partners are LAROMED GmbH (Schleswig), responsible for sensor hardware and system integration, and Fraunhofer IMTE (Lübeck), responsible for user-centred design and evaluation. GUARDIAN is funded under the DATIpilot Module 2 (BMBF) programme and runs from April 2026 to March 2028.
Technologies & Application Areas
- Large Language Models (LLM Integration)
- Human Activity Recognition (HAR)
- Multimodal Sensor Data Fusion
- Speech-to-Text & Text-to-Speech
- Context-Adaptive Interaction Logic
- Privacy-by-Design & Privacy-Compliant AI Architecture
- IoT & Edge Computing
- Medical Technology & Ambient Assisted Living (AAL)
- AI-driven Process Automation
