Generative AI Agent for Legal Case Analysis Support

Overview

Domestic legal professionals face excessive workloads due to a surge in cases, leading to trial delays and inefficient legal services. By introducing AI-based case analysis and natural language Q&A, repetitive administrative tasks can be automated, allowing legal experts and courts to focus on critical resolution. A generative AI agent-based service is essential to improve the speed and accessibility of legal services.

Project description

The project established an AI dataset by refining over 4.54 million legal records, including precedents, statutes, and commentaries, to design specialized language models for legal documents and Q&A. These core functions enable the automatic analysis of 8,000 pages per hour and provide natural language Q&A based on case records. By utilizing large-scale legal data and AI technology, the system serves as a decision-support assistant for legal professionals, drastically streamlining the heavy task of analyzing massive case files.

Key technological features

• LLM: Solar-Legal-LNC: Based on Upstage Solar-mini (10.7B), optimized for the Korean language and legal domain

• Model Training Pipeline:

  • Preprocessing: Text Cleaning (NLP) and Optical Character Recognition (OCR)
  • Domain Adaptation: Continuous Pre-training (Further Pre-training)
  • Post-training: Supervised Fine-tuning (SFT) and Instruction-Finetuning
  • Preference Alignment: Applied DPO (Direct Preference Optimization) for model alignment

• Advanced RAG: Integration of Fine-tuning and Graph RAG (Entity-Relationship Extraction & Knowledge Graph Construction)

Expected outcomes

The service significantly boosts the operational efficiency of legal experts by reducing work time by over 20% and increasing the number of cases processed by more than 15%. Beyond these productivity gains, it enhances the overall quality of legal services and improves accessibility for the public, thereby contributing to the promotion of national legal welfare.

The developer, Law & Company Consortium, successfully validated the legal case analysis service through a pilot program with the law firm Hwawoo in 2024. Building on this achievement, the consortium plans to officially expand the service to various law firms and legal entities in 2025 by enhancing the system's performance and developing additional specialized features.

Launch year: 2024

Attribution

Case study courtesy of the National Information Society Agency, Republic of Korea, Apolitical's Content Partner. Sourced from Use Cases of Public AI Service, Vol.2, https://eng.nia.or.kr/site/nia_eng/ex/bbs/View.do?cbIdx=31975&bcIdx=29182&parentSeq=29207