Tool Demonstration of SofIA-AI: Generating Use Cases and Classes from Interviews with LLMs

Abstract

Requirements elicitation through stakeholder interviews remains common in software engineering, but transforming unstructured conversations into formal artifacts (e.g., use cases or UML models) is costly and prone to ambiguity and information loss. Large language models (LLMs) can support this task, although their use raises challenges related to traceability, reproducibility, and incorrect or unsupported outputs. This paper presents SofIA-AI, an LLM-based extension of SofIA, an industry-oriented methodological ecosystem grounded in model-driven engineering. Starting from an interview transcript, the approach semi-automatically extracts actors, use cases, and domain elements and generates UML models while preserving a human-in-the-loop principle, so that the analyst validates intermediate artifacts before consolidation. The paper emphasizes the end-to-end workflow, analyst interaction for quality control, and methodological integration to maintain traceability across the SofIA lifecycle.