The use of large language models (LLMs) in education, research, and decision-making support is growing. However, users are unable to understand how they operate internally, which makes trust and openness difficult to achieve. Current Explainable Artificial Intelligence (XAI) methods usually offer explanations after the fact for models that predict things. These approaches frequently fail to produce content since these systems analyses information using a sequential, probability-driven process. This paper introduces Dynamo AI, an XAI framework with layers that aims to make generative intelligence systems more trustworthy. It does this by putting explainability right into the process of generating content. The framework adopted organizes explainability into four layers: model output, structured ways of showing reasoning, explanations that people can understand, and optional cues for checking information. This study outlines the architecture of Dynamo AI and presents a brief user study that examined users' trust in the system, their satisfaction levels, and the additional time required to receive a response. The outcome of the study was that users trusted and liked Dynamo AI more than standard generative outputs, with only a small increase in response time and incorporating layered explainability offers a practical path toward improving transparency in generative AI systems.
Anish Krisna Sudhakar (Thu,) studied this question.