Dinotesia is working with SBS to advance a large-scale AI search system for a broadcast archive. The companies said on March 20 they are upgrading semantic, multimodal search based on about 100,000 hours of broadcast video data.

The upgrade aims to go beyond simple keyword matching and enable users to find desired scenes using combined conditions such as people, actions, backgrounds and situations. The collaboration comes as competition after the spread of generative AI shifts from model performance to data infrastructure and how it is used.

Dinotesia is also reviewing whether it can secure fast search processing and scalability using its multimodal vector search technology and a VDPU (Vector Data Processing Unit), an acceleration chip for vector data processing. The VDPU accelerates vector operations at the hardware level to process large volumes of video data. In large-scale archive environments, key factors include not only search accuracy but also data processing speed and bandwidth efficiency.

SBS is applying an in-house AI search system to broadcast video across genres including news, current affairs and educational programming, and is conducting comprehensive tests of search accuracy, response speed and ease of use. It is also gradually applying its in-house AI-based facial search technology to reflect demand for person-centric searches, and said it plans to detail ways to apply the system across broadcasting operations including production, editing and archive management.

Dinotesia CEO Moo-kyung Jung (정무경) said, "Going forward, what will matter more than the model itself is how quickly large-scale data can be searched and connected to actual use."

The collaboration is being 추진 on the basis of a national R&D task, "Development of a vector DB for long-term memory storage of hyperscale AI models," within the SW computing industry source technology development programme overseen by the Ministry of Science and ICT and the Institute of Information and Communications Technology Planning and Evaluation (IITP).