Performance Improvement of LLMs for Regulatory Document Understanding based on Modified RAG Approach

A regulatory-document RAG system that foreshadows my current focus on agent-readable domain memory.

A modified retrieval augmented generation framework for regulatory documents. This is one of the applied roots of my current interest in turning domain documents into agent-readable memory.

Performance Improvement of LLMs for Regulatory Document Understanding based on Modified RAG Approach

problem

Regulatory documents are difficult for general LLMs because answer quality depends on retrieving the right provision, preserving context, and making the source trail visible.

key idea

Modify the RAG pipeline around regulatory-document structure so the model answers with stronger source grounding and less brittle context selection.

my role

Lead author; designed the applied RAG framing and evaluation direction.

methods

• Retrieval-augmented generation
• Regulatory document chunking
• LLM answer evaluation

evidence / results

• Published in JKIIS
• Established the applied-RAG root of later legal/tax and institutional assistant projects

why this belongs in the portfolio

• Shows how domain documents can become a queryable substrate
• Connects retrieval design to downstream answer reliability

authors

Jae-Hyun Baek, Jon-Lark Kim

venue / status

JKIIS

Published work; ResearchGate profile used as public metadata reference.