Fig. 2

Agentic workflow for scenario generation with user feedback loop. This figure illustrates the detailed structure of the agentic AI workflow implemented in n8n for healthcare simulation scenario design. The process begins with an input comprising “user parameters + research output,” derived from the initial interaction agent’s retrieval-augmented generation (RAG) process. This input is fed into a sequence of specialized LLM agents (LLM 1 through LLM 7), each responsible for a specific sub-task of scenario generation (decomposition-sub-agent task). Prompt chaining ensures that the output of each agent informs the subsequent agent’s input, maintaining contextual coherence. The workflow demonstrates parallelization, where three separate agentic workflows generate supplementary materials (ABGs, Lab Exams, Imaging) concurrently with the main scenario components. An iterative refinement loop, involving a “scenario reviewer” agent (LLM 6) and a “scenario editor” agent (LLM 7), ensures quality control and formatting. The “final output” (complete scenario + supplement materials) is then presented back to the user for review, editing, and final approval before implementation in simulation-based education