Retrieval-augmented generation (RAG) systems often suffer from treating documents as isolated pieces of data, lacking the ability to understand connections between them. To address this, developers have explored graph RAG approaches but faced challenges with complexity. The open-source Graph RAG Project and GraphRetriever propose a simpler solution by combining existing vector search with metadata-based graph traversal, allowing connections to be defined at runtime without needing graph construction or storage. A case study using a Rotten Tomatoes dataset demonstrates how GraphRetriever enriches AI-generated responses by connecting unstructured reviews with structured movie metadata through metadata fields like movie_id and genre as graph edges. This approach enhances the retrieval process, enabling the system to provide more contextual and comprehensive answers to natural language queries about movies. By dynamically traversing these metadata connections, GraphRetriever bridges the gap between subjective review content and objective metadata, improving the depth and relevance of responses without requiring a pre-built knowledge graph. The system can respond to queries with a wider context, enhancing the AI's ability to generate intelligent and complete answers, as shown in a demo where it successfully retrieves and contextualizes movie reviews for family movies, demonstrating the practical application of GraphRetriever in AI search and recommendation systems.