Retrieval-augmented generation (RAG) is the standard for integrating private or domain-specific knowledge into large language models. However, most RAG systems still produce hallucinations because the retrieval step is flawed. Large language models can only process the information they receive, and if the retrieved passages are inadequate, the model will invent unsupported information. Therefore, the accuracy and trustworthiness of a RAG system largely depend on its retrieval capabilities. This article explores relationship-aware retrieval as a solution to these retrieval weaknesses. It introduces RudraDB-Opin, a free, relationship-aware vector database, as a practical implementation. RudraDB-Opin is designed for learning, prototyping, and real-world projects. It can handle a significant number of vectors and relationships. This capacity allows for the modeling of substantial knowledge bases. The database aims to demonstrate various retrieval patterns discussed in the article.