1. Overview (Project Background)

This project was initiated to address the massive document writing burden and quality variance issues that arise in GC Biopharma's Annual Product Quality Review (APQR) process.
APQR is a critical quality task that requires regular preparation of reports spanning hundreds of pages based on data dispersed across multiple systems (LIMS, QMS, ERP, etc.), requiring significant time and resources for data collection, review, and organization.

Accordingly, GC Biopharma initiated the project with the goal of building a system that leverages AI technology to streamline the quality document writing process and secure accuracy and consistency in document quality.

2. Challenge (Problem Definition)

Prior to project initiation, the quality document writing process had the following structural limitations.

• Massive document volume and repetitive work burden
  To prepare annual APQR reports spanning hundreds of pages, data had to be collected, processed, and reviewed from multiple systems, and repetitive manual work created excessive workload.

• Human error and result variance
  Manual data entry and review processes had high error potential, and document quality and result consistency varied depending on individual staff capabilities.

• Data dispersion and utilization inefficiency
  Quality-related data was dispersed in silo form across multiple systems such as LIMS, QMS, and ERP, making integrated analysis and utilization difficult.

3. Solution (Resolution Approach)

GC Biopharma and Megazone Cloud built an AI-based quality document writing support system and implemented data integration, document automation, and accuracy enhancement in phases.

• Data platform construction
  A data platform was configured to collect, store, process, and analyze quality-related data from QMS, LIMS, ERP, etc., establishing a foundation for stable integrated management of data needed for document writing.

• Prompt Engineering and Rule-based writing
  Prompt engineering was applied for document summarization and key information extraction, and numerical calculations, format specification, and domain-specific logic were supplemented with Rule-Based Coding to ensure accuracy and reliability of results.

• SQL-based Context Augmentation
  Through predefined SQL, necessary data was selectively retrieved and the corresponding context was delivered to the LLM, minimizing hallucination and implementing responses based on internal data.

• Phased construction and enhancement approach
  The project was conducted in the form of PoC → Phase 1 construction → Phase 2 expansion, with prompts and features continuously improved based on operational feedback.

4. Result (Achievements)

Through the construction of the AI-based quality document writing support system, the following tangible achievements were accomplished.

• Significant reduction in document writing time
  Time spent on data collection, processing, analysis, and draft writing was reduced by over 2,845 hours, significantly alleviating staff workload.

• Secured document quality consistency and accuracy
  By generating documents based solely on system data, consistent quality output was reliably secured and human error was effectively prevented.

• Secured user satisfaction and expansion potential
  Positive feedback was received from actual users regarding document standardization, reduced missing items, and improved data inquiry convenience, and expectations for expansion to additional quality documents were secured.

• Implemented factory-specific customized report automation
  A system was established to automatically generate APQR and related reports for each factory (Ochang, Eumseong, Hwasun).