I build AI systems that stay capable under real-world constraints.

My work spans GPU/NPU heterogeneous systems, real-time intelligence, large language models, and human–computer interaction, with a focus on making AI responsive and dependable on edge and on-device platforms.

Interested in research collaborations on efficient AI systems, LLM serving, and practical deployment.

About

I am Taemin Kim, an M.S. student in the Department of Computer Science at Korea University. I study how to deploy and serve AI models efficiently when computation, memory, power, and latency are limited.

My long-term goal is to make advanced AI more reliable, accessible, and useful outside the lab: on devices, in time-sensitive systems, and in products people can actually use.

Research interests

GPU/NPU heterogeneous systems Real-time systems Large language models Edge & on-device AI Human–computer interaction

Recent projects

• VLM Interrupt-Based Robot Navigation SystemKorea University · 2025

  A vision-language-model-based robot navigation project. Details are limited by an LG NDA.

• Prompt Modularization & Emotion Recognition with LLMKAIST · 2024

  A modular prompting approach for emotion and mental-health recognition, developed during an Interactive Computing Lab internship.

• Hansung Blossom GPT Serving2024

  An LLM-serving project focused on practical deployment and system engineering.

• Multimodal LLM to Multimodal GenerationETRI & Hansung University · 2024

  Research extending multimodal language models toward multimodal generation.

See all projects, experience, and awards

Recent publications

• CoRT: Supporting Hard and Soft Real-Time Tasks in Batched LLM ServingICCAD 2026

  Yeongwoo Ha*, Taemin Kim*, Changhun Han, Chan Heo, Jaeheon Kwak, Hyosu Kim, Kilho Lee, Hoon Sung Chwa, and Sangeun Oh†.

• Self-Abstraction Learning for Effective and Stable Training of Deep Neural NetworksarXiv 2026

  Wonyong Cho*, Taemin Kim*, Jungmin Kim, Jeong Rae Kim, and Sung Hoon Jung.

• TipMate: An Explainable Fair Tip Recommender System based on LLMIEEE Access 2026

  Taemin Kim, Junhyuk Seo, Hyejung Ko, Keejun Han, and Woonghee Lee†.

• Zero-Shot Voice Cloning Based Emotion-Preserving Video DubbingIEIE Journal 2026

  Junhyuk Seo*, Taemin Kim*, Hyejung Ko*, and Heeseok Oh†.

See all publications

Get in touch

I welcome conversations about efficient AI systems, real-time model serving, edge intelligence, and research collaboration. Reach me at taemin6697@gmail.com.