Systems for scalable AI
We build full-stack systems for scalable training, high-performance inference, and reliable deployment of foundation models across diverse computing environments.
Menu
Scalable &
Efficient AI Lab.
KAIST AI
We develop algorithms and systems that make advanced AI efficient, scalable, and reliable. Our mission is to make foundation models and intelligent agents practical across a wider range of computing environments.
About SEAL
We study how to make advanced AI systems efficient by design.
Modern AI capabilities increasingly depend on large models and significant computational resources. Our research asks how new algorithms, model architectures, and systems can reduce these costs while preserving—or improving—model capability.
We work across the AI stack, spanning model architectures, training and inference systems, compression and deployment, and reasoning and agentic systems. Our goal is to develop principled methods that make foundation models practical at larger scales and in more practical settings.
Research
Current research directions
View all publicationsEfficient foundation models
We study model architectures, compression, and long-context methods that reduce the cost of training and inference without sacrificing capability.
Reasoning and agents
We investigate efficient reasoning, planning, inference-time scaling, self-improvement, and tool use for agents operating over long horizons.
Prospective students
Interested in joining SEAL?
We welcome applications from graduate students and research interns interested in efficient AI, large language models, and AI systems.
Application information