Enterprise software environments generate substantial volumes of technical documentation that are inherently large, fragmented, and difficult to navigate. In the context of Warehouse Management System (WMS) platforms, developers, consultants, and support teams must routinely locate precise information under operational time pressure. Conventional keyword-based retrieval mechanisms prove inadequate in this setting, as they lack the semantic understanding required to bridge the gap between natural-language queries and heterogeneous technical corpora. This thesis proposes and evaluates a multi-source Retrieval-Augmented Generation (RAG) system designed to ground language model outputs in verified, domain-specific documentation. The architecture employs hierarchical parent-child chunking with LLM-generated summaries, organized across source-scoped Qdrant collections.