Telepix's 6 Key Technologies Adopted at Space AI Conference

**Telepix Shines Research Capabilities at Earth Observation AI Workshop Jointly Hosted by NASA and ESA** Telepix, a South Korean space artificial intelligence (AI) company, garnered attention for its outstanding research achievements at a prestigious international workshop co-hosted by the National Aeronautics and Space Administration (NASA) and the European Space Agency (ESA), leaders in global space development. At the 'International Workshop on AI Foundation Models for Earth Observation' held from the 5th to the 7th in Frascati, Italy, at ESA's Earth Observation Centre (ESRIN), Telepix revealed its unique presence among Korean companies with a total of six research achievements accepted, including one oral presentation. This event was a forum to discuss the latest development trends and achievements of foundation models, a core AI technology in the field of Earth observation satellite imagery. Particularly, its significance is even greater as this marks the first time NASA and ESA have jointly hosted an event on this topic. Foundation models are the latest AI technology that pre-trains on vast amounts of unstructured data to autonomously identify complex patterns, and can then be applied to various tasks with only a small amount of additional data. They hold revolutionary potential in the field of Earth observation, where acquiring large-scale labeled training data is challenging. Jeong Young-sang, team leader of Telepix's AI Solutions Team, successfully introduced 'SatChat,' a large language model (LLM)-based satellite imagery utilization chatbot applying multi-agents, during the oral presentation session. 'SatChat' is an innovative system designed to allow non-experts to intuitively search and utilize satellite imagery and related information, and has already proven its commercial-level technological capability through supply contracts with domestic public institutions since its official launch last December. Concurrently, in the poster session, five in-depth research achievements were presented, including methodologies to maximize the accuracy of raw material classification in satellite imagery under limited data environments, and efficient adaptation strategies for foundation models utilizing LLM agent technology. Kwon Da-rong-sae, head of Telepix's Data Science Division, commented on this achievement, stating, "It is very meaningful that Telepix's research achievements were numerously selected as the only Korean company at an international conference co-hosted by NASA and ESA, which lead global space development." He added, "These research achievements will be immediately applied to our AI-based satellite utilization solutions, such as 'SatChat' and 'MetalScope,' a satellite information-based raw material logistics analysis service, significantly contributing to customers obtaining desired information more accurately and efficiently." Telepix has already been consistently recognized for its unique technological prowess and innovation in the global space AI market, having won the Best Paper Award at the 'International Conference on Learning Representations (ICLR),' the most prestigious academic conference in AI, and with 'MetalScope' being selected as the first Korean company to be featured as a satellite-based Earth observation use case by the World Economic Forum (WEF).