Chanwoo Park
Seoul, South Korea
I’m Chanwoo Park. My work focuses on building pre/post training datasets that unlock various capabilities of large language models. I am also interested in building AI solutions for financial domain, with end-to-end solutions in consideration. With various experience of optimizing DNN training in GPU clusters, I am always open to various research opportunities that face computational challenges.
news
| Jan 07, 2026 | SK Telecom released A.X K1, a 519B parameter MoE LLM that outperforms DeepSeek-V3.1 in math and coding benchmarks. As an intern at the Omnimodal Foundation Model Office, my work contributed to enhancing the model’s reasoning capabilities. |
|---|---|
| Oct 29, 2025 | From November 2025, I will start research internship at SK Telecom! Hope we have great experiences with the Omnimodal Foundation Model Office!  |
| Oct 29, 2025 | Our papers, “Beyond Line-Level Filtering for the Pretraining Corpora of LLMs” and Ko-MuSR: A Multistep Soft Reasoning Benchmark for LLMs Capable of Understanding Korean are now available on Arxiv! |
| Jul 02, 2025 | Our Korean-specialized LLM research was featured in DigitalToday! We developed Llama-Thunder-LLM, Thunder-Tok tokenizer (44% token reduction), and a Korean benchmark. |
| Jun 27, 2025 | Our paper, “UDC-VIT: A Real-World Video Dataset for under-Display Cameras” has been accepted to ICCV 2025! |