FusionRAG-Ex : an Interactive RAG Framework with retrieval and confidence-aware explanations

RAG improves factual grounding in LLMs, yet diagnosing whether outputs are faithfully supported by retrieved evidence remains challenging. We present FusionRAG-Ex, an interactive framework for explainable and faithfulness-aware RAG analysis. FusionRAG-Ex integrates retrieval relevance and token-level confidence into explanation generation, producing highlights that distinguish grounded from weakly supported responses. The system enables real-time inspection of retrieval scores, answer correctness, and explanation signals across retrievers, LLMs, and datasets, supporting systematic analysis and development of more trustworthy RAG systems.