Gunjun Lee

Research interests: Computer architecture, Machine learning, LLM, Speculative decoding, LLM inference / serving

Work Experience

• Manage server with pfSense, Ansible, NVIDIA DCGM, Prometheus, Grafana, OpenVPN

• Developed a Recommendation system based on Machine Learning for the in-app matching and deploying to production.
• Built model training / inference pipeline and data pipeline with NVIDIA Triton Inference Server, Pytorch, BigQuery, Kubeflow, Prometheus/Grafana and FastAPI.
• Optimized ML serving costs by reducing software overhead.
• Developed a iOS application based on CoreML and Metal-written custom kernel functions.

• Developed a Android / iOS application based on TensorFlow Lite.
• Developed a WebGL-based Application.

Education

Seoul National University, PhD in Department of Intelligence and Information 2025.09 - Current (Advisor: Jung Ho Ahn)

Seoul National University, MS in Department of Intelligence and Information 2024.03 - 2025.08 (Advisor: Jung Ho Ahn)

Seoul National University, BS in Electrical and Computer Engineering 2015.03 - 2022.02

Publications

• From Tokens to Layers: Redefining Stall-Free Scheduling for LLM Serving with Layered Prefill (MLSys, 2026)
• Gunjun Lee, Jiwon Kim, Jaiyoung Park, Younjoo Lee, Jung Ho Ahn
• Embed GitHub
• keywords & skills: C++, CUDA, vLLM, torch.compile, CUDA graph, ZeroMQ, Tensor Parallelism, torch Extension, nsys, ncu
• The New LLM Bottleneck: A Systems Perspective on Latent Attention and Mixture-of-Experts (arXiv, 2025)
• Sungmin Yun, Seonyong Park, Hwayong Nam, Younjoo Lee, Gunjun Lee, Kwanhee Kyung, Sangpyo Kim, Nam Sung Kim, Jongmin Kim, Hyungyo Kim, Juhwan Cho, Seungmin Baek, Jung Ho Ahn