MediCare-Agent: A Multi-Agent AI System for Post-Discharge Nephrology Patient Care Using LangGraph, RAG, and Large Language Models

The MediCare-Agent system demonstrates the transformative potential of multi-agent AI in the domain of post-discharge nephrology patient care. By leveraging LangGraph, FAISS-indexed RAG, Groq-hosted Llama 3.3 70B, SQLite patient management, and DuckDuckGo real-time search, the project achieved its goal of building a clinically structured, personalized, and fully zero-cost post-discharge support system. Features such as patient-context query enhancement, structured triage classification, mandatory medical disclaimers, and a three-step hierarchical routing strategy effectively address the limitations of general-purpose chatbots in clinical patient-facing applications. The system delivers GPT-4-class response quality at 3–4 seconds end-to-end latency with no recurring infrastructure cost — a property critical for adoption in resource-constrained healthcare settings.   Future enhancements will focus on integration with HL7 FHIR-compliant Electronic Health Record (EHR) systems for real-time patient record access, replacing the current static SQLite database. A formal randomized controlled clinical trial with real nephrology patients will be conducted to obtain quantitative accuracy, patient satisfaction, and health outcome metrics. Multi-source RAG augmentation incorporating PubMed abstracts, KDIGO clinical guidelines, and FDA drug label data will broaden knowledge coverage. Streaming token responses will reduce perceived latency. Voice interface support will improve accessibility for elderly patients with limited digital literacy. A longitudinal analytics dashboard tracking patient symptom trends and medication adherence will provide clinicians with population-level post-discharge insights.

[1] GBD Chronic Kidney Disease Collaboration, "Global, regional, and national burden of chronic kidney disease, 1990–2017: A systematic analysis for the Global Burden of Disease Study 2017," The Lancet, vol. 395, no. 10225, pp. 709–733, Feb. 2020. [2] Shortliffe, E. H. and Cimino, J. J. (2014). Biomedical Informatics: Computer Applications in Health Care and Biomedicine, 4th Edition, Springer. [3] Lee, J. et al. (2020). BioBERT: A Pre-trained Biomedical Language Representation Model for Biomedical Text Mining. Bioinformatics, vol. 36, no. 4, pp. 1234–1240. [4] Singhal, K. et al. (2023). Large Language Models Encode Clinical Knowledge. Nature, vol. 620, pp. 172–180. [5] Touvron, H. et al. (2023). Llama 2: Open Foundation and Fine-Tuned Chat Models. arXiv preprint arXiv:2307.09288. [6] Lewis, P. et al. (2020). Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks. Advances in Neural Information Processing Systems (NeurIPS), vol. 33, pp. 9459–9474. [7] Zakka, C. et al. (2024). Almanac — Retrieval-Augmented Language Models for Clinical Medicine. NEJM AI, vol. 1, no. 2. [8] Johnson, J., Douze, M., and Jégou, H. (2021). Billion-Scale Similarity Search with GPUs. IEEE Transactions on Big Data, vol. 7, no. 3, pp. 535–547. [9] LangChain Inc. (2024). LangGraph: Building Stateful, Multi-Actor Applications with LLMs. Documentation v0.2.28. Available: https://langchain-ai.github.io/langgraph [10] Reimers, N. and Gurevych, I. (2019). Sentence-BERT: Sentence Embeddings Using Siamese BERT-Networks. Proceedings of EMNLP, pp. 3982–3992.