CGM-Led Multimodal Tracking with Chatbot Support: An Autoethnography in Sub-Health

TL;DR

This paper tackles the lack of CGM use in sub-health populations by presenting a six-week autoethnography that combines continuous glucose monitoring, multimodal sensing, and an LLM-driven chatbot to support everyday health management. Using a retrieval-augmented generation pipeline and vision-language diet analysis, the study demonstrates how CGM data can surface hidden glycemic dynamics, while conversational feedback translates data into small, actionable adjustments, fostering gradual lifestyle change. It introduces the Continuous Experience Monitoring (CXM) framework and discusses benefits and ethical considerations of AI-powered health companions in preventive care, emphasizing cultural sensitivity and platform portability. While limited to a single participant, the work advances HCI and personal informatics by reframing CGM as a first-class signal for everyday well-being and reflective practice.

Abstract

Metabolic disorders present a pressing global health challenge, with China carrying the world's largest burden. While continuous glucose monitoring (CGM) has transformed diabetes care, its potential for supporting sub-health populations -- such as individuals who are overweight, prediabetic, or anxious -- remains underexplored. At the same time, large language models (LLMs) are increasingly used in health coaching, yet CGM is rarely incorporated as a first-class signal. To address this gap, we conducted a six-week autoethnography, combining CGM with multimodal indicators captured via common digital devices and a chatbot that offered personalized reflections and explanations of glucose fluctuations. Our findings show how CGM-led, data-first multimodal tracking, coupled with conversational support, shaped everyday practices of diet, activity, stress, and wellbeing. This work contributes to HCI by extending CGM research beyond clinical diabetes and demonstrating how LLM-driven agents can support preventive health and reflection in at-risk populations.

Figures (3)

• Figure 1: User-end data collection and demo of user interface. The first author tracked personal health using CGM and Apple Watch (a–d). Additional measures including blood pressure, SAS anxiety scores, and body weight were manually entered into the Health app (A) (e, g). Dietary intake was logged via photos and conversations with the chatbot (f). All data, including chatbot interactions, were uploaded at the same frequency as CGM to a private MongoDB database secured with AES-256 encryption.
• Figure 2: Backend application workflow of the CGM–chatbot system. The client uploads multimodal health data, including continuous glucose monitoring (CGM) values, every five minutes to the server. The server hosts a retrieval-augmented generation (RAG) module pre-trained on metabolic disease knowledge. When abnormal glucose levels are detected outside of sleep periods, a chatbot dialogue is triggered. Users can also upload meal photos, which are processed by Qwen-VL to estimate calories and glycemic index (GI). Both physiological signals and dietary information are passed to the intelligent agent, which integrates them with the Tencent Hunyuan text generation system to produce context-sensitive feedback. The response is then returned to the user as real-time, lifestyle-linked health guidance.
• Figure 3: Six-week health metrics overview. Left: Longitudinal trends of BMI (kg/m²), systolic blood pressure (mmHg), Self-Rating Anxiety Scale (SAS) scores, average sleep duration (hours), and heart rate variability (HRV, ms). Across the six weeks, BMI, blood pressure, and SAS scores decreased, while sleep duration and HRV increased, indicating improved physiological and psychological regulation. Right: Summary of CGM-derived metrics, including mean glucose, estimated HbA1c, time in range (TIR), excursions above and below range, and coefficient of variation (CV). These metrics remained largely within healthy ranges, highlighting the visibility of lifestyle impacts without evidence of insulin resistance.