AI/ML Researcher at Kakao Corp., working on LLM pre-training. Personally diving deep into alignment (SFT, RLHF/RLVR) and modern architectures.

• M.S. in Artificial Intelligence, Seoul National University
• B.S. in Biology and Business Administration, Seoul National University

• LLM Pre-training @ Kakao Corp.
• Developing the Kanana model family — Kakao's bilingual LLMs
• Building large-scale pre-training pipelines on JAX/TPU and PyTorch/GPU

• Pre-training at scale — µP, hyperparameter transfer, scaling laws
• Architectures — Mixture-of-Experts, Linear attention (DeltaNet, Gated DeltaNet), MLA
• Alignment — SFT, RLHF, RLVR, GRPO

• Kanana-2 언어모델 훑어보기
  Instruct.KR 2025 Dec Meetup — Agents, Seoul, Dec 2025 · [event] — co-presented Pre-training of Kakao's MoE LLM
• Featured in 요즘IT
  Speaker interview ahead of Instruct.KR Dec 2025 — on Kanana team's MoE development journey
• 데이터는 없지만 LLM은 학습하고 싶어 — Code, Math 데이터 개발기
  if(kakao) 25, Sep 2025 · [video] — data pipeline development for LLM pre-training (code & math corpora)
• Kakao's journey with JAX and Cloud TPUs
  Google Cloud Blog, Aug 2025 — co-authored technical deep-dive on Kakao's JAX/TPU adoption for Kanana pre-training
• 국내 최초 MoE 모델 'Kanana-MoE' 개발기
  Kakao Tech Blog, Jul 2025 — co-authored development story of Korea's first open-source MoE LLM
• Building and Serving the Next Generation AI Models with JAX
  Google Cloud Next '25, Las Vegas, Apr 2025 · [session]

• Kanana: Compute-efficient Bilingual Language Models
  Kanana LLM Team, arXiv preprint, 2025
• Clinical Note Owns its Hierarchy: Multi-Level Hypergraph Neural Networks for Patient-Level Representation Learning
  Nayeon Kim et al., ACL 2023 (Oral)

• ✉️ lana.ny@kakaocorp.com