Multi-Modal Fact-Verification Framework for Reducing Hallucinations in Large Language Models