NAVER | D2 STARTUP FACTORY

In conventional Transformer models, the key-value cache (KV cache) grows with context length, limiting their ability to handle complex tasks on memory-constrained devices. TraceWeave addresses this challenge with Phase-TITAN, an alternative to the Transformer architecture, and proprietary AI models built on it. By keeping memory usage constant even as context grows, Phase-TITAN enables advanced AI agents to run entirely on-device, without relying on the cloud. Founded by researchers with backgrounds at ETH Zurich, Google, and KAIST, TraceWeave is targeting applications in mobility, robotics, drones, and smart glasses.