AI Research Engineer at DeepAuto.ai and a combined M.S./Ph.D. candidate at KAIST, I have pioneered scalable, efficient attention mechanisms for large language models, including the award-winning HiP and Delta Attention algorithms. My work has led to a 52% reduction in serving costs and enabled million-token context inference on a single GPU, with multiple first-author publications at top-tier conferences like ICLR. My unique strength lies in bridging theoretical innovation with practical system optimization, delivering real-world impact in AI deployment. Looking ahead, I am committed to advancing the frontier of efficient AI systems, making state-of-the-art language models more accessible and sustainable at scale.

• Joined as CAIO from Aug. 2025. Lead agentic engineering AI projects.

• Developed ScaleServe, a cost-efficient LLM serving framework that reduces end-to-end serving costs by approximately 52% by integrating novel, training-free attention mechanisms.

• Invented HiP Attention (ICLR 2025), a training-free attention algorithm that speeds up long-context inference by 50% and enables serving million-token contexts on a single GPU via KV cache offloading.

• Designed Delta Attention, a novel correction algorithm that boosts sparse attention accuracy by 20-30% on the RULER benchmark with only a marginal (<10%) latency overhead.

• Engineered and integrated custom attention modules into serving frameworks like vLLM and SGLang, reducing computational complexity for long contexts from quadratic (O(n^2)) to near-linear (O(n)).

• Combined M.S./Ph.D. program

• GPA: 3.97/4.3

• College of Engineering Dean's List (Spring 2022)

• College of Engineering Leadership Award on Research Excellence (Spring 2022, Spring 2023)

• Python

• C++

• C#

• PyTorch

• Hugging Face

• vLLM

• SGLang

• OpenAI Triton

• .NET

• Authors: Jeffery Willette, Heejun Lee, Sung Ju Hwang

• arXiv Preprint

• Authors: Heejun Lee*, Geon Park*, Jaduk Suh*, Sung Ju Hwang

• arXiv Preprint

• Authors: Heejun Lee*, Geon Park*, Youngwan Lee*, Jaduk Suh*, et al.

• ICLR 2025

• Authors: Jeffrey Willette, Heejun Lee, Youngwan Lee, Myeongjae Jeon, Sung Ju Hwang

• ICLR 2025

• Authors: Heejun Lee, Jina Kim, Jeffery Willette, Sung Ju Hwang

• ICLR 2024

• Authors: Heejun Lee, Minki Kang, Youngwan Lee, Sung Ju Hwang

• ICLR 2023