We Were Already Building It: A Development Organization's Real-World AI Story (2025 Edition)

SK Planet is focusing not on developing its own LLM, but on improving product value, boosting development productivity through AI Coding, and innovating the way it collaborates with AI.

Examples already built include Oggeul Bot, which automatically generates the first comment on the OK Cashbag community, and AI Moment, which combines images, voice, and copy to create advertising videos. An in-house automatic AI newsletter delivery feature and RB Dialog, consisting of a chatbot builder and consultation solution, are also in operation.

What's currently being built spans a wider range. DMP/CDP Seg Lens, which recommends customer segments from natural language and images, uses GPT-4.1-nano, GPT-4o-mini, and business data optimization design, while Insight Lens, an internal text2sql assistant, helps generate queries using Claude 4 Sonnet, GPT-5, Llama 3.3 70B 6bit, and Multi-stage RAG.

In the consultation area, automatic responses to 1:1 inquiries automate draft writing and quality improvement, and in the advertising area, the Intelligent ADN Configuration System optimizes exposure weights and minimum unit prices in real time. MAB is used in this process, and GMM is also being tested recently to address its limitations.

Projects still at the lab stage include target population expansion based on Proximity ML, TTS & Speech2Sing, and Real-time Voice AI. Research continues on voice and song generation, real-time voice processing, and prediction based on regional data.

The conclusion is clear. While competing on ultra-large models carries high cost and risk, data quality, domain-specific RAG, prompt engineering, and operational feedback loops become cumulative assets. SK Planet has chosen a strategy of focusing on services and data rather than models, building sustainable competitiveness on top of replaceable LLMs.