Skip to content
Jaeyun Kim

Blog

Notes, mostly on serving models and the plumbing around them.

2026

DPA-MoE accepted to Findings of EMNLP 2026

August 1, 2026 · research, moe, inference, quantization

Our runtime precision control framework for MoE inference was accepted to Findings of EMNLP 2026 — my first paper, and one I led end to end.

Add+i enters full operation at ADD

July 1, 2026 · add+i, llm, service

Add+i, the air-gapped full-stack generative AI platform I built at the Agency for Defense Development, moved from beta into full operation.

2025

2024

Joined KRAFTON's Deep Learning Division

October 1, 2024 · career, llm

Started as a programming intern at KRAFTON, splitting time between LLM service development and on-device model serving for inZOI.