Enterprise-Grade Test Case Generation Framework Combining Retrieval-Augmented Generation with Multi-Modal Requirement Analysis

Abstract

The creation of software test cases demands significant engineering effort and often results in incomplete coverage
and limited traceability to requirements. This study presents a comprehensive framework designed to automate the generation of test cases from diverse requirement sources, including PDF documents, user interface images, and unstructured text
descriptions. The proposed system utilizes Retrieval-Augmented Generation methodology, incorporating domain-specific knowledge repositories to guide the generation process. By combining the GPT-4o language model with ChromaDB vector storage and
Lang Chain workflow management, the framework implements a multi-dimensional quality assessment mechanism with adjustable
acceptance criteria. Experimental validation conducted using representative test scenarios from web application domains demonstrates the framework’s effectiveness. The system generates test cases in under 2 minutes per scenario, achieves approximately 90% coverage of explicit requirements, and maintains semantic consistency with professional test standards. Bidirectional traceability is established through automated requirement identifier mapping. The RAG-based approach reduces unsupported assertions compared to standard language model prompting without knowledge base grounding. The system provides export functionality in multiple formats, including Behavior-Driven Development specifications, IEEE 829 standard reports, and common data exchange formats, facilitating integration with established test management platforms such as TestRail, Jira, and Azure DevOps. The architecture supports both collaborative web-based usage and standalone desktop deployment through PyWebView technology