Graph-Grounded LLMs: Leveraging Graphical Function Calling to Minimize LLM Hallucinations