[E-Daily Reporter Shin Yeong-bin ] CrowdWorks, Inc.(355390)has been awarded a contract to build training data aimed at enhancing artificial intelligence (AI)’s ability to understand Korean context and perform logical reasoning. The project will create high-complexity reasoning data based on public documents that combine visual elements—such as tables, graphs, and diagrams—with text.

CrowdWorks, Inc. announced on the 7th that it had been awarded the “Composite Document-Based Knowledge Inference Data” construction task as part of the “2026 AI Hub Training Data Inference Project,” which is organized by the Ministry of Science and ICT and implemented by the National Information Society Agency (NIA).

CrowdWorks, Inc. will lead this project as the lead organization, working in a consortium with WOT.CO.,LTD. and Zendive.

The AI Hub Training and Inference Data Project was designed to build high-complexity AI inference data applicable to industrial settings. The government plans to enhance the inference capabilities of domestic AI models and establish a data foundation capable of understanding Korean-based domain knowledge and industrial contexts.

Among the 10 projects in total, the consortium is responsible for building knowledge inference data based on complex documents. The consortium plans to collect and refine public documents in fields such as public administration, science and technology, and laws and patents to produce more than 10,000 high-quality inference training data sets.

The data is based on multimodal documents that combine visual elements—such as tables, graphs, and diagrams—with text. The goal is to build a complex reasoning dataset that goes beyond simple question-and-answer data to understand context from multiple angles and derive causal relationships.

CrowdWorks, Inc. will apply its data design capabilities—which structure AI’s thought processes—to this project. It will analyze areas of inference weakness across various fields and design complex thought flows and high-difficulty questions using inference techniques such as Chain of Thought (CoT), Thought Tree (ToT), and Thought Graph (GoT).

The company explains that this will enable AI to go beyond simply identifying a single correct answer and instead learn the thought processes humans go through when solving complex problems.

Data quality control will also be strengthened. To reduce hallucination, clear sources will be matched to each inference stage, and a hybrid verification pipeline combining automatic verification by large language models (LLMs) with expert review will be implemented. The plan is to ensure data integrity through this process.

CrowdWorks, Inc. believes that the high-difficulty inference data built through this project will contribute to improving the inference performance of Korean-language AI models. The company explains that this could serve as a foundation for AI to solve complex business processes and technical challenges in industrial sectors such as manufacturing, finance, and the public sector.

A CrowdWorks, Inc. official stated, “Upon completion of this project, we will contribute to securing proprietary AI technology that understands Korean-based administrative and industrial contexts, as well as data sovereignty, amid the onslaught of global Big Tech companies,” adding, “Based on our expertise in data design and processing, as well as our quality management capabilities accumulated across various industries, we will contribute to strengthening the competitiveness of the domestic AI industry.”