LLM‑guided HRV biofeedback and CBT‑style support inside the Lief app.


The Lief device provides heart rate variability (HRV) biofeedback, the ECG patch sensor hapticly sharing when real-time HRV dips below a target threshold.
While the biofeedback assists in increasing HRV and thus decreasing stress, human coaches are employed to help with lifestyle adjustments, which can be expensive and often not used.
An AI HRV coach is available 24/7 and has real-time access to the user's wearable metrics. Having an AI agent support users in conjunction with human support was a near-term goal for the product offering.
Lief Therapeutics wanted a chat AI that would be able to 1) support users with real-time access to their heart data, 2) be experienced in Cognitive Behavioral Therapy (CBT) methods to assist users in moving through personal challenges, 3) have basic user calendar integration to detect HRV trends, and 4) offer basic customer support.
I built an agentic system with tool use, leveraging the OpenAI API for rapid development and deployment. Tool use allowed real-time DB queries for user data. Collected CBT reference documents were chunked and made accessible via a vector store.
• Conversation and agent routing: customer service, coaching support, data retrieval & analysis
• RAG over CBT knowledge base; designed tone/style constraints and safety rails.
• Secure authenticated access to user device data, and correlating with calendar event insights
System evaluation performed on conversations, with variety of topics, replies, and errors.
Thorough evaluation of routing targets, as well as tone/style within a given target.
Errors were tracked and analyzed for further development efforts.