A Proposed Framework for Automating Hybrid Quantum-Classical Requirements Engineering Using Large Language Models: An Exploratory Pilot Study

Abstract

          Hybrid quantum-classical systems lack systematic support for requirements engineering (RE), a phase that is currently manual and requires scarce dual-domain expertise. This study investigates the feasibility of using Large Language Models (LLMs) for two core RE tasks: (1) classifying requirements as classical or quantum-amenable, and (2) generating draft mathematical formulations (QUBO). We propose a three-agent framework using GPT-4o and compare its performance against simple baselines on a controlled synthetic dataset of 120 requirements. The proposed framework achieved 90% overall classification accuracy on a held-out test set (n=40). A detailed error analysis reveals three distinct failure patterns: polysemy, implicit dependencies, and ambiguous scope. Cost analysis shows GPT-4o is 875× more expensive per call but requires no labeled data. LLMs show promise for automating hybrid RE tasks.