Article Impact Level: HIGH
Data Quality: STRONG
Summary of Frontiers in Digital Health https://doi.org/10.3389/fdgth.2025.1670464
Dr. Morris Gellisch et al.
Points
- LLMs excel in summarizing and reasoning but are limited by text-only inputs, reflecting subjective and delayed user reports.
- Integrating real-time physiological signals like HRV enriches LLM context, enabling adaptive and objective interactions in healthcare.
- A streamlined architecture routes real-time HRV data from wearable sensors directly into a generative AI environment.
- Core HRV metrics were derived in real-time and made accessible to OpenAI’s GPT models via REST and WebSocket endpoints.
- ChatGPT successfully accessed real-time HRV data, performed analyses, generated visualizations, and adapted feedback to autonomic shifts.
Summary
This research introduces a novel, modular pipeline designed to establish a real-time interface between physiological biomarkers and large language models (LLMs). Recognizing the inherent limitations of LLM inputs to user-generated text—which can be subjective, delayed, and consciously filtered—the study emphasizes the critical value of integrating real-time physiological signals. This integration enables LLMs to incorporate objective autonomic state indicators alongside linguistic input, fostering more adaptive and context-sensitive interactions across various applications, including learning, decision-making, and healthcare settings. The architecture specifically focuses on routing real-time heart rate variability (HRV) data from wearable sensors directly into a generative AI environment.
The methodological approach utilized a validated heart rate variability sensor to decode Bluetooth-transmitted R-R intervals via a custom Python script. From these intervals, core HRV metrics—including Heart Rate (HR), Root Mean Square of Successive Differences (RMSSD), Standard Deviation of NN intervals (SDNN), Low-Frequency/High-Frequency (LF/HF) ratio, and the percentage of successive normal sinus R-R intervals that differ by more than 50 ms (pNN50)—were derived in real time. These continuously computed values were then made accessible to external applications, such as OpenAI’s GPT models, through their publication via REST and WebSocket endpoints, supported by a FastAPI backend.
The established pipeline successfully demonstrated a live data flow from autonomic input to conversational output, enabling a language model to respond to real-time physiological shifts in natural language. In several proof-of-concept scenarios, ChatGPT was shown to access real-time HRV data, perform descriptive analyses, and generate visualizations. Crucially, the model adapted its feedback dynamically in response to autonomic shifts induced by varying cognitive loads (low and high), confirming the feasibility and utility of integrating real-time physiological data for enhanced, context-aware AI interactions.
Link to the article: https://www.frontiersin.org/journals/digital-health/articles/10.3389/fdgth.2025.1670464/full
References
Gellisch, M., & Burr, B. (2025). Establishing a real-time biomarker-to-LLM interface: A modular pipeline for HRV signal acquisition, processing, and physiological state interpretation via generative AI. Frontiers in Digital Health, 7. https://doi.org/10.3389/fdgth.2025.1670464
