Writer 1 / Study Program : Arkan Alexei Andrei / Computer Science Writer 2 / Study Program : Joshua Mihai Daniel Nadeak / Computer Science Writer 3 / Study Program : James Smith Wigglesworth / Computer Science Title : Comparative Analysis of Retrieval-Augmented Generation Architectures for Indonesian Law Counselor : Adila Alfa Krisnadhi, S.Kom., M.Sc., Ph.D. Alfan Farizki Wicaksono, S.T., M.Sc., Ph.D. The Indonesian legal system is characterized by its complex and fragmented documentation, posing challenges in information retrieval, accessibility, and contextual understanding. This study investigates Retrieval-Augmented Generation (RAG) systems, evaluating vector-based and graph-based retrieval mechanisms independently to address these challenges. By leveraging dense vector-based models like bge-m3 for semantic encoding and the LexID Knowledge Graph derived from over 1500 Indonesian legal documents for relational insights, this research explores advanced retrieval techniques integrated with Large Language Models (LLMs) to improve precision, contextual relevance, and efficiency in legal information processing. The methodology includes testing vector-based retrieval for robust semantic matching and graph-based retrieval for capturing relationships within interconnected legal texts. Fine-tuning LLMs for tasks such as legal question answering, summarization, and document interpretation is also explored. Metrics such as Precision@K, Recall@K, MRR, DCG@K, MAP, BLEU, ROUGE, METEOR, and BERTScore comprehensively evaluate the retrieval systems’ accuracy, relevance, and ranking quality. Results demonstrate that vector-based retrieval excels in semantic matching and high recall, while graph-based retrieval is particularly effective for tasks requiring contextual relationships and interdependencies. This research identifies best practices in prompt engineering and parameter optimization for LLMs tailored to the Indonesian legal context. By addressing systemic inefficiencies, it contributes to advancing equitable access to justice and supports the development of scalable, AI-driven legal informatics solutions.