International Journal of Science, Engineering and TechnologyAuthors: Rishu Burnwal, Aishwarya Verma, Saurabh Kumar, Abhishek Singh
Abstract: Healthcare systems worldwide are increasingly burdened by administrative inefficiencies, diagnostic delays, and fragmented patient data management. Traditional hospital informa-tion systems, while adequate for basic record-keeping, fail to leverage the transformative potential of Artificial Intelligence (AI) in clinical workflows. This paper presents MedAI, an end-to-end Intelligent Hospital Ecosystem implemented as a full-stack web appli-cation integrating a multi-provider AI orchestration layer supporting Google Gemini 2.0 Flash, Groq LLaMA 3.3 70B, OpenRouter GPT-4o Mini, and NVIDIA LLaMA 3.1 70B with a graceful provider-agnostic fallback chain. The core clinical AI modules include: MedAssist (conversational LLM-based triage chatbot with emergency keyword detection and multilingual support), MedReport (multimodal medical report analyzer via Google Gemini vision), PharmaSafe (drug interaction checker with severity clas-sification using structured JSON pipelines), and ClinicalIQ (clinical decision support generating ICD-10-coded differential diagnoses with SOAP note summarization). The operational layer encompasses JWT-authenticated role-based access control, patient management with paginated search, appointment scheduling, live analytics dashboards with Recharts visualizations, staff and department management, billing, laboratory test ordering, and WebSocket-based real-time synchronization. The system was evaluated using 39 Playwright E2E tests across three browser engines, achieving 100% pass rate with zero WCAG 2.1 AA accessibility violations. Performance benchmarks demonstrate sub-200ms API response times for database operations and graceful fallback under AI provider unavailability. MedAI demonstrates that a well-architected, lightweight hospital AI ecosystem can be built and deployed using modern open-source tools, of-fering a practical roadmap for AI-assisted healthcare delivery in resource-constrained environments.
