en.Wedoany.com Reported - Crowdworks has won a South Korean government AI reasoning data project to address bottlenecks in Korean language model capabilities at industrial sites. The company announced on the 7th that it has been selected for the "Compound Document Basic Reasoning Data" construction task under the "2026 AI Hub Learning Data (Reasoning) Project," supervised by the Ministry of Science and ICT and promoted by the National Information Society Agency (NIA). The task is executed as a consortium led by Crowdworks, with Wort Intelligence and Zendive as co-participants.
Recent evaluations of Korean language expertise in domestic and global generative AI models show that most models have not reached the average human level in reasoning ability. This is due to limitations in models' ability to understand context from multiple perspectives and derive high-level causal relationships, a problem considered a core challenge hindering AI application in industrial settings. The government initiated this project to overcome such technical limitations, build high-difficulty reasoning AI data applicable to industrial sites, and enhance the quality competitiveness of the domestic AI industry.
The consortium will collect and refine multimodal public documents combining visual elements such as tables, charts, and drawings with text, producing over 10,000 high-quality reasoning learning data items, thereby constructing a Korean multimodal compound reasoning raw dataset. Crowdworks' strategy is to go beyond simple data collection and focus its data engineering capabilities—which structure AI thinking processes—on this task. By analyzing the weak reasoning links in fields such as public administration, science and technology, laws, and patents, the company will precisely design complex thought processes and high-difficulty problems using reasoning techniques such as Chain of Thought (CoT), Tree of Thought (ToT), and Graph of Thought (GoT).
To prevent AI hallucinations, the project matches clear sources at each reasoning stage and applies a hybrid validation pipeline combining Large Language Model (LLM) automatic verification with expert review to ensure data integrity. Crowdworks anticipates that if high-difficulty reasoning data is successfully constructed, the Korean language reasoning performance of AI models that have faced bottlenecks will approach human levels. This will also help AI take the lead in solving complex business processes and technical challenges that have previously been hindered in various industrial fields such as manufacturing, finance, and public services.
A Crowdworks representative stated that upon completion of this project, it will help secure independent AI technology and data sovereignty for understanding Korean administrative and industrial contexts. The company will contribute to strengthening the competitiveness of the domestic AI industry based on its years of experience in data design, processing, and quality management accumulated across various industrial sectors.
