Establishing a real-time biomarker-to-LLM interface: a modular pipeline for HRV signal acquisition, processing, and...

Large language models are capable of summarizing research, supporting clinical reasoning, and engaging in coherent conversations. However, their inputs are limited to user-generated text, which reflects subjective reports, delayed responses, and consciously filtered impressions. Integrating physiological signals provides a clear additional value, as it allows language models to consider real-time indicators of autonomic state alongside linguistic input, thereby enabling more adaptive and context-sensitive interactions in learning, decision-making, and healthcare. Therefore, we present a streamlined architecture for routing real-time heart rate variability data from a wearable sensor directly into a generative AI environment.

Using a validated heart rate variability sensor, we decoded Bluetooth-transmitted R-R intervals via a custom Python script and derived core heart rate variability metrics (HR, RMSSD, SDNN, LF/HF ratio, pNN50) in real time. These values were published via REST and WebSocket endpoints through a FastAPI backend, making them continuously accessible to external applications-including OpenAI's GPT models.

A live data pipeline from autonomic input to conversational output. A language model that does not just talk back, but responds to real-time physiological shifts in natural language. In multiple proof-of-concept scenarios, ChatGPT accessed real-time HRV data, performed descriptive analyses, generated visualizations, and adapted its feedback in response to autonomic shifts induced by low and high cognitive load.

This system represents an early prototype of bioadaptive AI, in which physiological signals are incorporated as part of the model's input context.