Crowdworks announced on the 16th that it shared the latest AI technology trends and introduced its data preprocessing technology at the 'AI Technology Briefing Session' hosted by Jae-Cheol Kim, AI Graduate School of the Korea Advanced Institute of Science and Technology (KAIST) held at COEX on the morning of the 16th.
This technology briefing session was designed to introduce the core source AI technology that KAIST is researching to the industry and the general public, and to promote AI technology diffusion and industry-academic cooperation. This event was held as part of the '2025 International Artificial Intelligence Expo (AI EXPO KOREA),' and Crowdworks participated in the lecture at the invitation of KAIST.
Yang Su-yeol, CTO of Crowdworks, gave a lecture on the interesting topic, “Why can’t AI read Manager Kim’s report that the CEO reads well?”
CTO Yang explained, “Although AI can read general document formats, it is still not easy to understand the ‘meaning’ contained in the document and extract it as metadata.” He continued, “Visual elements such as charts and diagrams must go beyond simple explanations and consider the context of surrounding sentences and paragraphs to configure meaning-based metadata so that AI can accurately retrieve related information and improve response quality.”
He continued, “Since our country’s documents have their own unique style and structure, using foreign parsers as they are will result in many errors,” and emphasized, “We need to reflect these characteristics of domestic documents and implement high accuracy through precise parsing and processing of tables and visual elements.”
Along with this, the company also introduced its own solution, 'Alpy Knowledge Compiler', which can preprocess various unstructured documents into a form suitable for RAG (Retrieval-Augmented Generation). The solution performs LMM (Large Multimodal Model)-based analysis on tables, charts, images, etc. in documents, and adds semantic metadata to improve search precision and query response quality. In particular, it can systematically analyze document structures by applying its own evaluation index that can quantify the complexity of documents for the first time in the industry, thereby reducing the possibility of data preprocessing errors and efficiently managing manpower and budget.
- See more related articles
You must be logged in to post a comment.