This paper addresses the problem of building a specialized semantic search engine for documents collected in small or medium-sized enterprises. It presents the results of a project that brought together computer scientists and entrepreneurs for the purpose of providing a common perspective regarding the implementation in company practice of a search engine based on the Polish version of Word-Net semantic relations. The core functionality of the search engine module is provided along with a discussion on how to arrange semantic similarity structures so as to ensure the efficient generation of relevant search engine results. Some patterns and similarity coefficients for hyperonymy, hyponymy, holonymy and meronymy relations are presented and analyzed for the purpose of producing relationship structures. Finally, the architecture of the system that can be implemented in a company is outlined.