Advancing Rule Development from NDAA Text through Integrated LLM and Machine-Based Reasoning Tools

Abstract

This research builds upon prior efforts by the authors to streamline Defense Federal Acquisition Regulation Supplement (DFARS) rule development from National Defense Authorization Act (NDAA) text using AI-based tools. The current research effort focuses on developing a unified, web-based interface that integrates previously developed prototypes and incorporates advanced Natural Language Processing (NLP), Large Language Models (LLMs), and Machine-Based Reasoning (MBR) techniques to improve automation in identifying, extracting, and recommending regulatory language changes. The proposed unified system connects modules for document ingestion, keyword and context identification, text summarization, clustering, and visualization—through an integrated backend and user interface, which will enable DPCAP staff to move seamlessly from NDAA review to DFARS draft generation. The tool also proposes novel Machine-Based Reasoning techniques that leverage LLM models for updating proposed rule language and summarizing public comments from Regulations.gov. The combined framework is being deployed on a secure, sponsor-accessible server and will be evaluated against real DFARS updates in collaboration with DPCAP subject matter experts. This work aims to significantly reduce manual analysis time, enhance the traceability of regulatory updates, and strengthen the Department of Defense’s capacity to apply AI responsibly in acquisition policy modernization.