This paper presents a systematic approach to improving the interpretability of Convolutional Neural Networks (CNNs), focusing on the VGG16 architecture, through the application of Layer-wise Relevance Propagation (LRP). Our objective is to enhance the transparency and explainability of deep learning models by visualizing the contribution of individual image pixels in the network’s decision-making process. We propose a custom implementation of LRP integrated with the VGG16 model, supported by robust preprocessing pipelines and real-time visualization capabilities via live video input. A detailed literature review on Explainable Artificial Intelligence (XAI) and LRP frames the foundation for our methodology. The resulting tool—LRP Detective—provides dynamic, pixel-level insights into model behavior, contributing to more transparent and trustworthy AI systems. This work holds practical value for researchers and developers aiming to deploy interpretable CNNs in real-world applications.