We propose a modular and scalable architecture for building explainable recommendation systems that leverages knowledge graphs, translational embeddings, and LLMs. First, enriched knowledge graphs are constructed, followed by embedding-based relationship capture. Rather than a post-hoc addition, explanations are integrated into the core recommendation process. Visual explanations highlight relevant paths in the knowledge graph that connect users to recommended items, while natural language explanations, generated through an LLM, provide intuitive justifications that users can easily understand. A MovieLens 1M case study, enhanced with DBpedia, demonstrates strong experimental results and qualitative evidence of coherent, justifiable recommendations, particularly in sparse data scenarios. This highlights the potential of our approach for improving explainable recommender systems.