Abstract
The advent of increasingly powerful language models has raised expectationsfor language-based interactions. However, controlling these models is achallenge, emphasizing the need to be able to investigate the feasibility andvalue of their application. We present PROMISE, a framework that facilitatesthe development of complex language-based interactions with informationsystems. Its use of state machine modeling concepts enables model-driven,dynamic prompt orchestration across hierarchically nested states andtransitions. This improves the control of the behavior of language models andthus enables their effective and efficient use. We show the benefits of PROMISEin the context of application scenarios within health information systems anddemonstrate its ability to handle complex interactions.