<![CDATA[Context-aware vendor recommendation remains a critical challenge in dynamic event planning systems, particularly in multilingual and temporally sensitive markets such as Indonesia. This paper presents MAMRS (Multi-Agent Multimodal Recommendation System), a novel graph-based architecture that integrates semantic similarity, temporal availability, and user interaction history within a heterogeneous knowledge graph processed using Graph Attention Networks (GAT). The system uniquely combines CSR-aware attention mechanisms with a local LLM (Mistral-7B) to deliver explainable, sustainability-focused recommendations while preserving data privacy. MAMRS introduces three key innovations: (1) A GAT-based reasoning layer that dynamically weights vendor relevance using both structural relationships and corporate social responsibility (CSR) scores. (2) A fusion scoring engine that optimizes for semantic, temporal, behavioral, and sustainability constraints (achieving 93.4% CSR compliance in output, validated through grid search and rule-based scoring), (3) A locally deployed LLM that generates natural-language justifications with 18.7% higher BLEU scores, complemented by ROUGE-L (0.44) and METEOR (0.39), and validated through a small-scale Likert-scale user study (n=5). Evaluation on 250+ real user queries and 1,200+ vendor profiles from BuatEvent.id demonstrates that MAMRS achieves: (1) 21.3% improvement in top-3 accuracy over dense retrieval baselines (with p < 0.05, two-tailed t-test). (2) 1.6s average latency (1.9s p95) on local infrastructure. (3) 88.7% recall@5 in cold-start vendor scenarios, including new vendors with sparse metadata. This paragraph of the first footnote will contain the date on which you submitted your paper for review, which is populated by IEEE. It is IEEE style to display support information, including sponsor and financial support acknowledgments, here rather than in an acknowledgment section at the end of the article. The current implementation supports Indonesian and English queries, with future work planned for additional ASEAN languages and domains such as education and healthcare. These results position MAMRS as an effective solution for regulation-sensitive, multilingual event-planning platforms that require auditable, trustworthy, and scalable recommendation rationales.]]>
