Exploring Trust and Transparency in Retrieval-Augmented Generation for Domain Experts

The increasing adoption of Retrieval-Augmented Generation (RAG) systems in domain-specific applications raises critical challenges around trust and transparency, particularly for domain experts who possess deep subject knowledge but limited technical familiarity with AI systems. This paper investigates the design and evaluation of a user-centric RAG-based application tailored for financial equity research, incorporating features such as confidence indicators, source transparency, and user control mechanisms. Through a user study with 50 financial professionals, we found that while confidence scores provided useful context, they alone did not significantly enhance trust. In contrast, transparency features, including source attribution and highlighting document sections used in responses, substantially improved trust and user understanding. Participants also expressed a strong preference for greater control over source documents, such as filtering trusted reports and interacting with specific data. This work provides actionable insights for designing RAG systems that support trust, transparency, and control in high-stakes decision-making domains.