Enhancing Legal LLMs through Metadata-Enriched RAG Pipelines and Direct Preference Optimization

TL;DR

This paper introduces Metadata Enriched Hybrid RAG and Direct Preference Optimization to improve the grounding, safety, and reliability of legal language models, especially in handling long documents and ensuring accurate outputs.

Contribution

It presents novel retrieval and optimization techniques tailored for legal LLMs, addressing their limitations in long document understanding and safe response generation.

Findings

Enhanced retrieval accuracy in legal corpora

Reduced hallucinations and errors in generated legal texts

Improved safety through better refusal mechanisms

Abstract

Large Language Models (LLMs) perform well in short contexts but degrade on long legal documents, often producing hallucinations such as incorrect clauses or precedents. In the legal domain, where precision is critical, such errors undermine reliability and trust. Retrieval Augmented Generation (RAG) helps ground outputs but remains limited in legal settings, especially with small, locally deployed models required for data privacy. We identify two failure modes: retrieval errors due to lexical redundancy in legal corpora, and decoding errors where models generate answers despite insufficient context. To address this, we propose Metadata Enriched Hybrid RAG to improve document level retrieval, and apply Direct Preference Optimization (DPO) to enforce safe refusal when context is inadequate. Together, these methods improve grounding, reliability, and safety in legal language models.