Kolon Benit debuts Korea’s first full-stack AI system with local chip

Kolon Benit CI

Kolon Benit announced Wednesday that it has developed South Korea’s first integrated AI system combining a domestically produced AI semiconductor with a large language model (LLM), marking a key milestone in the nation’s push for sovereign AI.

According to IT industry sources, Kolon Benit recently unveiled its new AI-based safety control system, AI Vision Intelligence, at the Exaone Partners Day hosted by LG AI Research. The solution brings together Rebellions’ neural processing unit (NPU), LG’s Exaone 3.5 LLM, and Kolon Benit’s own monitoring software - forming a fully localized full-stack AI system, from chip to model to application.

Developed in collaboration with LG AI Research, the system is designed for construction site safety surveillance using AI-powered closed-circuit cameras (CCTV). By leveraging both graphics processing units (GPUs) and NPUs, the platform enables high-performance, real-time inference. Training is performed on GPU infrastructure while real-time inference is handled by NPU servers equipped with Rebellions’ ATOM AI chip, boosting both speed and accuracy.

Exaone 3.5 enhances the system by generating more detailed and context-aware natural language descriptions of detected hazards, improving situational awareness for on-site monitoring personnel.

Currently deployed at Kolon Global construction sites, AI Vision Intelligence detects a range of safety risks in real time - including missing safety helmets, unauthorized access to danger zones, absence of signal operators, and approaching heavy machinery.

The announcement aligns with the Lee Jae-myung administration’s broader goal of promoting sovereign AI technologies by supporting domestically developed chips, models, and applications.