An Interactive Tool for iterative Domain Elicitation and Metamodeling

Abstract

Building a metamodel from a natural language domain description is hard to get right in a single shot and hard to correct once the artifact is complete. Existing generation tools also assume modeling expertise on the user side, leaving non-modeling experts unsupported. We present MetaLoop, a web-based tool that builds a metamodel one concept at a time through a conversational interface. An interactive elicitation phase first lets the user and the agent agree on the concepts to model. The tool then generates the metamodel chunk by chunk. Each chunk is validated against a generated sample instance, explained in natural language, and submitted for user approval before being added to the final metamodel. The tool also supports attaching a reference document. This document grounds a coverage report that shows which elicited concepts are supported and which concepts were not captured during elicitation. The tool is implemented as a LangGraph agent exposing two callback points, human validation and user response, that decouple the elicitation logic from the Gradio frontend and make the agent reusable behind other interfaces. This paper describes the tool’s architecture and its interactive features.