Enhancing Explainable RAG through Neural-Hungarian Optimization : Hybrid Framework for Transparent Question Answering