A Multimodal Text- and Graph-Based Approach for Open-Domain Event Extraction from Documents — ThinkLLM