MediSentinelAI: A Safety-Critical Multi-Agent LLM Framework for Symptom Triage and Geospatial Healthcare Coordination

This paper describes MediSentinelAI as a reported PWA combining multimodal prescription extraction, four-tier symp-tom triage, offline red-flag matching, medication reminders, sensory alerts, and geospatial emergency coordination. The updated count-based dossier reports 79.2% exact complete-record accuracy on 250 prescription images, 91.7% triage ac-curacy with 45/45 observed critical-case recall on an assumed method mapping of 180 vignettes, and 97.6% notification success across 500 trials compared with 98.6% for the reported native FCM baseline. The PWA-native difference was 1.0 percentage point in delivery and 17.6 ms in aggregate median latency under the stated test, but the success deadline and raw logs remain unavailable. Results conflict with earlier headline summaries and are not independently verified. Future deployment requires reproducible evaluation records, verified privacy controls, and independent clinical and regulatory re-view. EHR integration using FHIR/HL7 and hospital bed-occupancy streaming remain future work.  

[1]     World Health Organization, Emergency Triage Assessment and Treat-ment (ETAT) Course: Manual for Participants. Geneva, Switzerland: WHO, 2005. [2]     N. Gilboy, T. Tanabe, D. Travers, and A. M. Rosenau, Emergency Severity Index (ESI): A Triage Tool for Emergency Department Care, Version 4. Rockville, MD, USA: Agency for Healthcare Research and Quality, 2012. [3]     R. M. Anjana et al., “Metabolic non-communicable disease health report of India: The ICMR-INDIAB national cross-sectional study,” The Lancet Diabetes & Endocrinology, vol. 11, no. 7, pp. 474–489, 2023. [4]     L. Masanneck et al., “Triage performance across large language models, ChatGPT, and untrained doctors in emergency medicine: Comparative study,” Journal of Medical Internet Research, vol. 26, Art. no. e53297, 2024, doi: 10.2196/53297. [5]     C. Y. K. Williams, B. Y. Miao, A. E. Kornblith, and A. J. Butte, “Evaluating the use of large language models to provide clinical rec-ommendations in the emergency department,” Nature Communications, vol. 15, Art. no. 8236, 2024, doi: 10.1038/s41467-024-52415-1. [6]     X. Tang et al., “MedAgents: Large language models as collaborators for zero-shot medical reasoning,” in Findings of the Association for Computational Linguistics: ACL 2024, Bangkok, Thailand, 2024, pp. 599–621, doi: 10.18653/v1/2024.findings-acl.33. [7]     World Health Organization, Ethics and Governance of Artificial Intelli-gence for Health: WHO Guidance. Geneva, Switzerland: WHO, 2021. [8]     Google, “Structured outputs,” Gemini API Documentation. [Online]. Available: https://ai.google.dev/gemini-api/docs/structured-output. Ac-cessed: Oct. 23, 2026. [9]     Google, “Gemini API deprecations,” Gemini API Documentation. [On-line]. Available: https://ai.google.dev/gemini-api/docs/deprecations. In-cludes Gemini 1.5 Flash shutdown information and current model routing; accessed Oct. 23, 2026. [10]    World Wide Web Consortium, “Service Workers,” W3C Recommenda-tion, Nov. 2022.