The use of Large Language Models (LLMs) in industrial settings has opened new possibilities for intelligent support within manufacturing systems. Nonetheless, standalone LLMs often struggle to reliably access specialized technical documentation needed in production environments. To address this, this thesis applies the development of a Retrieval-Augmented Generation (RAG) system focused to manufacturing that facilitates structured access to technical documentation and production-related information. Improving over existing RAG systems, this research presents an enhanced version that incorporates enriched metadata to improve document grounding and retrieval accuracy. A dedicated evaluation metrics program was also designed to measure and analyze retriever performance, allowing for systematic comparisons of different system versions.