Validating LLM-Generated Data Grounded in Technical Documents: An Application in Community Planning

Published

February 3, 2026

Author(s)

Juan Fung, Daniel Kofi Stephens, Alden Dima, Michael Majurski

Abstract

This report presents a novel approach to generating synthetic variations of technical community planning documents using Large Language Models (LLMs). We developed a comprehensive framework for transforming domain-specific resilience, adaptation, and sustainability (RAS) planning documents into more user-friendly versions while preserving semantic content. Our methodology employs multiple modification strategies including linguistic simplification, jargon removal, content augmentation, and tone adjustment, implemented through parallel processing pipelines. The system generates both individual and cumulative document modifications, which are then validated using a comprehensive framework that includes multiple similarity metrics. This validation framework provides a robust assessment of the semantic preservation achieved by the user-friendly modified documents. The process provides a template for generating and validating rich data while balancing user needs and maintaining technical accuracy in specialized domains.

Citation

Technical Note (NIST TN) - 2361

Report Number

2361

NIST Pub Series

Technical Note (NIST TN)

Pub Type

NIST Pubs

Download Paper

https://doi.org/10.6028/NIST.TN.2361

Local Download

Keywords

community resilience, synthetic data, large language models, validation

Citation

Fung, J. , Stephens, D. , Dima, A. and Majurski, M. (2026), Validating LLM-Generated Data Grounded in Technical Documents: An Application in Community Planning, Technical Note (NIST TN), National Institute of Standards and Technology, Gaithersburg, MD, [online], https://doi.org/10.6028/NIST.TN.2361, https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=960732 (Accessed February 17, 2026)