Abstract: FR-PO0027
Supporting Clinician Adoption of Hemodiafiltration: A Real-Time Artificial Intelligence (AI) Chatbot with Verified Clinical Sources
Session Information
- Artificial Intelligence and Digital Health at the Bedside
November 07, 2025 | Location: Exhibit Hall, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Artificial Intelligence, Digital Health, and Data Science
- 300 Artificial Intelligence, Digital Health, and Data Science
Authors
- Holik, Radek, Renal Research Institute, New York, New York, United States
- Yueh, Sheng-Han, Renal Research Institute, New York, New York, United States
- Bergling, Karin, Renal Research Institute, New York, New York, United States
- Usvyat, Len A., Renal Research Institute, New York, New York, United States
- Maddux, Franklin W., Fresenius Medical Care AG, Waltham, Massachusetts, United States
- Zhang, Hanjie, Renal Research Institute, New York, New York, United States
Background
Post-dilution high-volume hemodiafiltration (HDF) has been used internationally for the past few decades and is now becoming available in the United States. Given that HDF is new to American clinicians, successful adoption depends on tools that are transparent, user-friendly, and clinically trustworthy. To support clinicians in implementing this therapy, we developed the HDF Chat Bot, an AI-powered web-based assistant. This tool is based on the Stuard et al. (2025) monograph as its main knowledge base. Unlike generic AI chatbots, every response is linked to a verified source, ensuring full transparency and reliability.
Methods
The HDF Chat Bot was implemented with retrieval-augmented generation (RAG) using three vector databases: BookDB (Stuard et al. monograph), ReferenceDB (20 peer-reviewed publications on HDF) and VideoDB (Advanced Renal Education Program educational videos). All sources were split into one-page chunks, enriched with metadata, and embedded via the OpenAI embedding API.
The ResponseAgent consolidates the retrieved chunks with the query and the most recent chat dialogues history to generate a source-linked response. Every response includes precise page numbers of the monograph, manuscript links and video links to ensure full traceability. The information flow from the user query to the response is illustrated in Fig. 1.
Results
The integration of three specialized sources and a unified ResponseAgent ensures the delivery of accurate context coverage and source traceability. The web interface supports text and voice input, real-time text-to-speech playback, and displays precise citations.
Conclusion
The HDF Chat Bot is a novel, real-time, source-verified, RAG-based assistant designed to educate and support clinicians implementing HDF. A key benefit of this approach is its ability to prevent hallucinations. Future work will focus on clinical validation and on quantifying its impact on protocol adherence and patient outcomes.
Funding
- Commercial Support – Renal Research Institute LLC,