“Generative artificial intelligence (AI) is like a newly arrived “confident intern” at corporations.”
Cho Gyu-gon, CEO of Fasoo, said during the keynote speech at the 'FDI 2025 Symposium' held at the Fairmont Ambassador Hotel in Yeouido, Seoul, on the 22nd, “Generative AI answers everything but is not perfect, so it is essential to establish data governance first to reliably delegate tasks that are broken down into smaller pieces.”
Fasoo presented 'establishing data infrastructure and governance in advance' as a solution for corporations to safely utilize generative AI.
CEO Cho noted, “A new AI model emerges every 2.5 days on average, and by 2026, 80% of enterprise software will incorporate AI capabilities,” while pointing out that, even if GPU (graphics processing unit) prices have stabilized downward, investing more than half of the budget in large-scale projects for system infrastructure, such as servers, is excessive. He explained that public large language models (LLMs) have clear limitations regarding expense and privacy, making the limited use of public LLMs and hybrid adoption of small LLMs (sLLMs) more realistic.
Cho emphasized that DLP (Data Loss Prevention) and content verification are essential for all data generated and output by generative AI. He introduced the 'RAY' data lake and metadata management platform and the AI governance solutions 'AI DLP' and 'AI DHC', stating, “Without control at the input and output stages, it is difficult to realize the benefits of investing in private LLMs.” He added, “A modular architecture that allows easy replacement of new models should be established.”
Yoon Kyung-gu, Deputy Minister, introduced the private LLM 'Ellm'. 'Alum' is equipped with a base model of the Zema-3 with 12 billion parameters and provides functionalities such as retrieval-augmented generation (RAG), report automation, and multi-turn conversation optimization. Yoon mentioned, “Thanks to an open-source model that has already surpassed the average expert MMLU score (89.8), the sLLM acceleration rule of ‘doubling performance every three months’ is becoming a reality.”
He also explained, “With the spread of model context protocols (MCP) and agent-to-agent standards, generative AI has begun to handle real-world tasks such as hotel bookings and payments,” adding that “'Alum' connects internal systems like electronic approvals and document management to the MCP server, establishing the 'human-in-the-loop' feature, which allows human intervention at key decision-making stages.”
Fasoo also showcased generative AI-based work tools. 'Knowledge Master' is a chatbot that queries and answers internal knowledge in bulk, 'Librarian' provides customized searches by document type, and 'Scribe' is a tool for automatically creating business plans and research reports. 'Prompt Studio' offers an environment to visually design complex multi-step prompts, while 'RAG Benchmark' measures even hallucinations to verify quality during model replacement.
CEO Cho noted, “Large corporations have gaps in security investments, and small and medium-sized enterprises do not know where to start,” stating, “Real AI innovation is possible only if private LLMs and governance infrastructure are built on foundational elements such as backup, encryption, and vulnerability analysis.”