This paper explores a hybrid approach combining the strengths of traditional algorithms and Large-Language Model (LLM) powered Retrieval Augmented Generation (RAG) systems. While LLMs excel at generating human-like text and processing unstructured data, they can be unreliable for structured data extraction. To address the challenge of explaining complex business process models, we first use algorithms to extract key information. Then, a case-based RAG system generates textual descriptions. We investigate two methods: a) generating individual element descriptions and ordering them, and b) enhancing coherence through a harmonization step to produce a more unified explanation. A significant contribution of our work lies in the achievement of improved scalability. Unlike existing methods that often suffer from performance degradation when handling larger process models, our hybrid approach demonstrates a remarkable degree of consistency, maintaining accuracy even as the model complexity increases. Furthermore, we have successfully minimized the required context window for the LLM. By relying on the data extracted by the algorithms, we significantly reduce the amount of information the LLM needs to process simultaneously. This reduction in context window size not only improves efficiency but also mitigates the risk of information loss or “hallucinations” that can occur when LLMs are tasked with processing excessively large inputs. Therefore, our approach provides a trustworthy and robust solution for generating accessible explanations of complex business processes.