Abstract:
Digital legal research in Indonesia requires faster, more accurate systems to
support practitioners in navigating complex regulatory documents. Identifying relevant
statutes and drafting or reviewing legal documents remains time-consuming and
error-prone due to scattered laws and inconsistent formats. A multi-agent large language
model (LLM) system was developed using GPT‐4.1 and Gemini‐2.5‐pro‐flash within a
retrieval-augmented generation (RAG) framework. LangChain and meta‐prompting help
structure agent workflows. The system includes three modules: Regulation Retrieval (for
finding relevant articles and Pasal to a certain case or issue), Document Drafting (for
generating structured contracts), and Legal Review (for highlighting inconsistencies).
Law student reviewers confirmed that the Retrieval module consistently cites accurate
Pasal and returns relevant statutes. Drafted documents meet basic structural and linguistic
standards. This study demonstrates how multi-agent LLM architectures, combined with
prompt engineering and RAG, can efficiently support Indonesian legal workflows by
automating regulation search, document drafting, and review. These methods can
significantly reduce time and effort in legal practice.
