MANAGING AI-ENABLED KNOWLEDGE FOR LARGE-SCALE SOFTWARE ENGINEERING: A RETRIEVAL-AUGMENTED APPROACH

Abstract

Large-scale software engineering organizations continuously produce diverse and extensive knowledge artifacts, such as source code, technical documentation, issue tracking records, and architectural decision documents. Effectively managing and reusing this knowledge remains a longstanding challenge due to information fragmentation, rapid system evolution, and the inherent limitations of traditional keyword-based knowledge management systems. Although recent advances in neural language models have shown strong capabilities in natural language understanding and generation, their direct application in software engineering contexts is limited by insufficient domain grounding, reliance on outdated information, and a lack of traceability. To address these challenges, this paper proposes an AI-driven Knowledge Management System (KMS) based on a Retrieval-Augmented Generation (RAG) architectural approach tailored for large-scale software engineering environments. The proposed architecture combines semantic retrieval with generative reasoning to enable context-aware and grounded access to organizational knowledge across heterogeneous software repositories. By conditioning generated responses on retrieved, project-specific artifacts, the system enhances accuracy, transparency, and adaptability to evolving knowledge bases. The paper presents the architectural design, methodological framework, and qualitative case studies focused on developer onboarding and technical debt mitigation, illustrating the potential of retrieval-augmented architectures as a foundation for next-generation knowledge management systems in software engineering.