Application for visualizing the AI model explanation by leveraging open-source explainable AI techniques

The increasing complexity and "black-box" nature of artificial intelligence models present significant challenges to their trustworthiness and adoption. While Explainable AI (XAI) techniques have been developed to provide transparency into model decision-making, their practical application is often hindered by high technical barriers, requiring programming expertise and complex configurations. This limits their accessibility for both AI researchers, who face time-consuming implementation, and non-technical users, who lack the necessary skills. This thesis details the design, development, and validation of VisX.AI, an interactive web application created to bridge this gap. The primary objective is to provide a unified, no-code platform that integrates open-source XAI techniques (such as Grad-CAM and SHAP) to empower users to visualize explanations for their own uploaded AI models and datasets. The system is built upon a generic and extensible microservice architecture, utilizing the Strategy and Template Method design patterns to ensure modularity and facilitate future expansion. The platform supports multiple AI modalities, including Computer Vision, Natural Language Processing, and Tabular data processing. Validation of the application confirmed that it successfully meets its core functional requirements, generating meaningful visual explanations tailored to different tasks and models. The architecture was proven to be extensible, allowing new tasks and XAI methods to be integrated with minimal modification. Usability testing with stakeholders revealed that the application effectively lowers the barrier to entry for XAI, though it also highlighted the challenge of model uploading for non-technical users. While performance remains a trade-off due to the computational demands of explanation generation, the project demonstrates a step toward making AI explainability more accessible, fostering greater trust and transparency in AI systems.