SAGE: Sensor-Augmented Grounding Engine for LLM-Powered Sleep Care Agent
Researchers' new system grounds LLM responses in real-time sensor data to provide actionable sleep advice.
A team of researchers from Seoul National University and the University of Sydney has introduced SAGE (Sensor-Augmented Grounding Engine), a novel framework designed to power more effective and trustworthy AI sleep coaches. The system directly addresses the pervasive 'Data-Action Gap,' where users of wearables and health apps collect vast amounts of sleep data but struggle to interpret it and turn it into practical improvements. SAGE works by creating a unified, queryable layer that normalizes continuous streams of sleep, physiological, and activity data from various sensors.
This technical foundation enables two key modes of interaction. First, it supports selective, system-initiated monitoring, where the AI agent only triggers notifications when it detects meaningful deviations from a user's personal baseline, thereby combating alert fatigue. Second, it allows for user-initiated Q&A, where natural language questions are translated into precise database queries. This ensures every LLM-generated response about sleep quality or advice is directly grounded in specific time periods, metric comparisons, and the user's own historical data, moving beyond generic tips.
The research, accepted to the CHI 2026 conference, articulates a new design space for evidence-based digital health messaging. By tethering the reasoning of a large language model to a user's concrete, personal data history, SAGE aims to significantly enhance the personalization, traceability, and ultimately, the trustworthiness of AI health agents. It represents a move from chatbots that guess based on general knowledge to assistants that reason with your unique biometric context.
- SAGE creates a queryable time-series layer from continuous sleep, physiological, and activity sensor data.
- It reduces alert fatigue via system-initiated monitoring that only triggers on meaningful deviations from personal baselines.
- Enables natural language Q&A where questions are translated into executable DB queries, grounding all LLM responses in personal data.
Why It Matters
It transforms sleep data into actionable, personalized advice, moving AI health agents from generic chatbots to context-aware coaches.