About

Hello! I'm Kunhee Kim, a PhD student at KAIST. I am interested in improving reasoning and decision-making in large-scale AI systems through inference-time computation.

My research focuses on how techniques such as search, planning, and adaptive inference can enhance the capabilities of foundation models beyond what is learned during training. I aim to understand how test-time computation can make models more controllable, reliable, and efficient.

Previously, I worked on controllable and personalized generative models for visual content, focusing on how to adapt pretrained models to new concepts and styles with minimal supervision. I am particularly interested in extending these ideas toward more general adaptive and reasoning-capable generative systems.

I am always open to discussions and collaborations!

Publications

    Experience

    NAVER Cloud - Visual Generation Team (Resident)
    Oct 2025 -- Apr 2026

    • Led end-to-end development of large-scale generative models (video diffusion), including data pipeline, model design, and training.
    • Trained billion-scale models using distributed multi-node GPU systems.
    • Built and processed synthetic datasets for training generative models at scale.

    NAVER Webtoon - Generative Model Team (Intern)
    Aug 2022 -- Feb 2023

    • Worked on generative models for visual content creation.
    • Collaborated with multiple researchers on style transfer and diffusion models.

    Hobby Projects

    Outside of my research, I enjoy building small projects and tools, particularly around design, web development, and presenting research more effectively.

    • Academic Website
      Designed and developed this website from scratch using HTML/CSS/JS, with a focus on clarity, minimal design, and efficient presentation of research.
    • Research Group Websites
      Created a reusable template for academic group websites, which you are welcome to use for your own group: example page (code). This template was later adopted and customized by multiple research groups, including KAIST CVML and the KAIST Visual Intelligence Lab.