Sungju Kim

Currently working on reinforcement learning for LLM agents, with a focus on enhancing their software engineering capabilities. Previously, I have conducted research and development in NLP, dialogue systems, information retrieval, embeddings, and question answering.

Experience

NAVER Cloud, Tech Lead (2023.01 - Present)
Hyperscale AI, Seongnam, South Korea

• Contributed to HyperCLOVA X 32B THINK Project through model development and research for SWE Agents, including creating SFT and RL datasets, developing training environments for SWE Agent RL, and babysitting model training during the General and SWE Agent RL phases. 2025
• Contributed as one of the core contributors to the HyperCLOVA X THINK Project. 2025
• Led a team that developed a product and model similar to GitHub Copilot, which is used by internal developers at NAVER. 2023-2024
• Contributed as one of the technical leads to the HyperCLOVA X project. 2024
• Contributed to the enhancement of the coding capabilities for HyperCLOVA X, an LLM developed by NAVER Cloud. 2023-2024

NAVER, Tech Lead (2022.08 - 2022.12)
NAVER, AI Research Engineer, NLP (2018.07 - 2022.07)
Clova AI assistant team, Seongnam, South Korea

• Developed a state-of-the-art Neural Information Retrieval model using NAVER’s extensive internal text data resources, achieving unprecedented performance in zero-shot information retrieval. 2022.
• Contributed to develop a large-scale autoregressive language model, HyperCLOVA. My contributions included data collection, purification, and conducting experimental work on zero-shot, few-shot, and p-tuning. 2020-2021.
• Developed Pretrained Language Models for Korean and Japanese. 2020-2021.
• Developed Out of Domain Detection model for dialogue system using sentence-embedding model. 2020-2021
• Developed Natural Language Understanding model. 2020.
• Developed and implemented a suggestion model for ambiguous user queries, significantly enhancing the system’s capacity to interpret and respond to a broad spectrum of unclear queries. 2018-2019.
• Improved dialogue manager of dialogue system. 2018-2019.

Nomad Connection, AI Researcher (2016.02 - 2018.07)
AI lab team, Seoul, South Korea

• Modeling intent classifier and sentence encoder for AI chatbot. 2017-2018.
• Developed Keyword extractor and Sentiment classifier. 2016-2017.
• Developed ID card OCR model development for non-face-to-face authentication. 2017.
• Developed Sentiment classifier. 2016

Education

Kyungpook National University, Computer Science, M.S. (2014.03 - 2016.02)
Computer & Robot Vision Lab, Daegu, South Korea

Kyungpook National University, Biotechnology, B.S. (2009.03 - 2014.02)
Daegu, South Korea

Award

Ecothon, Grand Prize (1st place), Awarded by the Minister of Environment
Team Qualc, 2016

• IOT smart oil filter for life management of vehicle’s oil filter.
• News: Ingenious ‘Environment + ICT’ Startup Idea Competition Ecothon… Qualc Team Grand Prize (title translated)

Selected Publications

• HyperCLOVA X Team, …, SJ Kim et al, HyperCLOVA X THINK 32B, arXiv preprint arXiv:2601.03286, 2025.
• HyperCLOVA X Team, …, SJ Kim et al, HyperCLOVA X THINK Technical Report, arXiv preprint arXiv:2506.22403, 2025.
• KM Yoo, …, SJ Kim et al, HyperCLOVA X Technical Report, arXiv preprint arXiv:2404.01954, 2024.
• SJ Kim, SH Kim, J Park, KM Yoo, IH Kang, The Bi-Cross Pretraining Method to Enhance Language Representation, Proceedings of the 33th Annual Conference on Human and Cognitive Language Technology, 2021.
• BS Kim, …, SJ Kim et al, What Changes Can Large-scale Language Models Bring? Intensive Study on HyperCLOVA: Billions-scale Korean Generative Pretrained Transformers, Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, 2021
• SJ Kim, SB Suh, JS Park, SH Park, DH Jeon, SH Kim, KD Kim, IH Kang, Self-supervised Learning Method using Heterogeneous Mass Corpus for Sentence Embedding Model, Proceedings of the 32th Annual Conference on Human and Cognitive Language Technology, 2020.
• SJ Kim, WW Kim, YS Seol, IH Kang, Cross Gated Mechanism to Improve Natural Language Understanding, Proceedings of the 31th Annual Conference on Human and Cognitive Language Technology., 2019.
• SJ Kim and SY Park, Lane-Level Positioning based on 3D Tracking Path of Traffic Signs, in Proceedings of the 11th Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications - Volume 3: VISAPP, (VISIGRAPP 2016), 2016.

Patents

• Method, Apparatus and Computer Program for Providing Coding Environment Optimized for Each User KR Patent Application No. 1020230177976, Published as KR1020250088172A in 2025 DOI Link

• Method and System for Processing Unclear Intent Query in Conversation System US Patent No. US11403345B2, Granted in 2022 Google Patents Link