Two-Stage Framework for Personalized Sleep Guidance via SLMs
September 22, 2026
This framework uses a multi-agent LLM pipeline to generate reasoning trajectories from unannotated wearable data, which are then distilled into Small Language Models (SLMs). The resulting SLMs outperform general and medical LLMs in generating personalized sleep guidance through supervised fine-tuning and Best-of-N selection.
HOW THIS AFFECTS YOU
●
builderYou can use agentic distillation to turn unannotated sensor data into high-performing, small-scale specialized models.
●
healthYou can deploy lightweight, privacy-preserving sleep guidance models directly on end-user devices.