Strengthening the foundation for using generative AI on internal networks by structuring administrative documents
Supporting operational innovation and the advancement of public-facing services

Document AI company Lomin (CEO Ji-hong Kang) announced that it has completed the deployment of its document data processing solution, “Textscope Doc Parser,” for the “Pan-Government Hyperscale AI Common Infrastructure Implementation Project,” promoted by the Ministry of the Interior and Safety and the Ministry of Science and ICT and carried out by the Samsung SDS consortium.

This project is the government’s first AI common infrastructure service for internal networks, enabling central and local governments to jointly use various generative AI services even within internal administrative networks without security concerns, with the goal of embedding AI across government operations to improve policy planning and the quality of public services.

Recently, as the use of generative AI has expanded within the government, discussions have continued on how to convert public documents into data that AI can actually understand and use.

In particular, as concerns have been raised that simply converting Hangul Word Processor (HWP) documents—which account for a significant portion of public administrative documents—into PDFs may fail to properly convey the document’s structure and context, the importance of structuring data while preserving the original document’s format and meaning has been highlighted.

In this context, the Pan-Government Hyperscale AI common infrastructure project established a document-structuring preprocessing system so that AI can utilize public documents.

As a technology partner in the Samsung SDS consortium, Lomin applied Doc Parser to the preprocessing stage that structures and refines public administrative documents used by the “Pan-Government AI common infrastructure” and related services so that large language models (LLMs) can learn and understand them.

Doc Parser supplied by Lomin is a document layout analysis solution that extracts key elements such as text, tables, and images from various documents—including official documents, reports, and administrative forms—and recognizes the document’s layout and reading order to structure data in a way that preserves the original structure and context.

In particular, it is characterized by the ability to parse and structure Hangul Word Processor (HWP/HWPX) documents—which are widely used in Korea’s public and business environments—in their original form without converting them into image-based formats such as PDFs.

Through this, it is designed to prevent document formats from being damaged or additional manual work from occurring even in complex layouts frequently found in public documents, such as multi-column structures, boxed templates, and table-to-caption relationships.

Through this deployment, a preprocessing system has been established to refine and structure administrative documents used in key administrative and public-facing services—such as document drafting support and searches of laws and guidelines provided by the Pan-Government AI common infrastructure—into formats suitable for LLM application.

Ji-hong Kang, CEO of Lomin, said, “It is highly meaningful that Lomin’s document structuring technology has been applied to real administrative sites within a pan-government AI common infrastructure project,” adding, “We have once again demonstrated our technological competitiveness in the preprocessing area that turns public documents into data that AI can use immediately. We will continue to advance our technology so that we can contribute to the realization of a digital platform government.”

Meanwhile, Lomin is preparing to launch “Zixy,” a vision-language-model (VLM)-based Document AI SaaS platform that consolidates the implementation know-how it has accumulated to date.

Zixy plans to support mid-sized and small-to-medium enterprises with limited capacity for system implementation so they can easily use high-performance AI document processing in a cloud environment.