[Case Study] Dongwon Group | Advancing AI OCR for Unstructured Document Processing with Generative AI

🚀 Key Takeaways at a Glance

✅Built a generative AI-powered OCR system to process Dongwon Group’s diverse invoices and unstructured documents.

✅Enabled flexible processing of new document formats and unstructured data that were difficult to handle with conventional OCR.

✅Implemented a workflow in which AI automatically recognizes and extracts key fields and table data after document upload, followed by user review.

✅Doosan Digital Innovation BU (DDI) enhanced core AI OCR capabilities, including generative AI infrastructure, prompt design, document recognition logic, and data extraction and validation logic.

✅Delivered LLM performance comparison for operational use, generative AI extraction module development, SSO integration, and code-system expansion.

✅Established a foundation to improve document processing scalability and reduce repetitive data checking and entry tasks.

Company

Dongwon Group is a comprehensive food company that has grown from its roots in the fisheries industry. The company has strong capabilities across global marine resource development, food production, and distribution. Drawing on long-standing expertise and accumulated technology, Dongwon Group provides a stable supply of high-quality seafood products and food ingredients while contributing to healthier, more abundant lifestyles through sustainable use of marine resources and continuous innovation in the food industry.

Challenge

Dongwon Group faced limitations in processing diverse document formats and unstructured data using conventional OCR alone.

Need to Process Unstructured Documents

Dongwon Group needed a more flexible way to handle unstructured documents and new invoice formats for which templates had not been predefined.
Operational Burden When Adding New Document Formats

If every new document format requires separate configuration or development, the operational burden on business teams increases.
To address this, Dongwon Group wanted a structure in which administrators could register frequently used templates and users could define extraction fields according to each document.
Manual Review and Data Re-entry Workload

Manually reviewing documents and transferring data to Excel or other tools is time-consuming and creates room for errors.
The company also needed higher levels of automation when working with keywords, table data, multi-page documents, and multiple files.
Need for Accuracy Improvement and Operational Management

AI OCR should not stop at extraction. It also needs to capture feedback on results and use that feedback to improve accuracy over time.
The project therefore included a feedback input function designed to support future accuracy enhancement.

Solution

DDI combined generative AI with OCR to implement an end-to-end AI OCR workflow that supports document type expansion, field recommendation, data extraction, result review, and Excel download.

[Data Processing Architecture Based on Key AWS Services]

Applied a GPT OCR Processing Architecture Based on Key AWS Services
- Amazon API Gateway: Served as the API call channel connecting OCR Web with AI OCR analysis functions.
- AWS Lambda: Executed OCR analysis requests, data processing, and AI OCR extraction after document upload.
- Amazon S3: Stored uploaded document files and data used for AI OCR processing.
- Amazon RDS: Managed operational data such as user information, template information, and OCR analysis results.
- Amazon Bedrock: Supported generative AI document analysis and GPT OCR extraction using the Claude 4 model.
- Amazon SES: Provided the foundation for system notifications and email delivery integration.
- Amazon EC2 / ELB: Supported the OCR Web operating environment and external access infrastructure.
Integrated the existing OCR system with a generative AI-based GPT OCR architecture.
Designed a hybrid cloud integration architecture.
Applied SSO and user authentication.
Implemented template-based AI OCR management features.
Applied a prompt-based field extraction configuration framework.
Automated extraction field recommendations powered by AI.
Implemented a PDF upload-based GPT OCR registration process.
Designed a confidence score-based extraction result review screen.
Established a feedback-based quality improvement framework for AI OCR.
Implemented administrator statistics and operational monitoring features.
Customized user screens to improve business usability.

DDI combined Dongwon Group’s existing OCR environment with an AWS-based AI OCR processing architecture and Amazon Bedrock-powered generative AI analysis. As a result, Dongwon Industries can manage unstructured documents and diverse invoices through templates and prompts, while operating an extensible GPT OCR workflow that includes AI recommendations, confidence-based review, and feedback management.

Benefit

With the generative AI-powered OCR system, Dongwon Group expanded its document processing coverage while improving response speed for new formats and overall business efficiency.