Retrieval-Augmented Generation (RAG) systems for enterprise technical documentation face a persistent vocabulary challenge: users refer to the same product by many different names, abbreviations, and legacy terms, while the underlying search index uses a controlled product taxonomy. This mismatch degrades retrieval precision and recall, ultimately reducing answer quality. I present DOC, the production conversational assistant deployed on docs.netapp.com, and describe two complementary techniques that leverage curated product-name knowledge artifacts to close this vocabulary gap. First, a query-time product-name normalizer rewrites user queries by mapping informal or ambiguous product mentions to their canonical names and appending the latest version identifiers. Second, a product-scope filter dynamically constructs structured filter expressions from a product catalogue, ensuring that retrieval is restricted to the set of documented products. Together, these lightweight, maintainable components yield measurable improvements in retrieval relevance without retraining any model. On a stratified sample of 300 real production queries from docs.netapp.com, with slug-derived ground-truth relevance judgments, curated normalization improves Mean Reciprocal Rank by +42%, +31%, and +24% over raw queries on BM25 and on two Azure dense retrievers (text-embedding-3-small and text-embedding-3-large ), all with non-overlapping 95% bootstrap CIs. A factorial ablation reveals that the Product-Scope Filter, when driven dynamically by the same detected products, contributes a complementary +25-78% MRR gain on top of every retriever/normalizer pair, and that the curated normalizer still out-performs a hyperparameter-tuned NPMI log-mining baseline trained on 93, 028 production query-URL pairs across both filter states. I share system design details, experimental results, and operational lessons from two years of production deployment spanning 60+ enterprise storage and cloud products.