Youngin Cho
KRAFTON · Research Engineer
About Me
I am a Research Engineer at KRAFTON, building embodied AI agents for games such as PUBG Ally. I earned my Bachelor’s degrees in Computer Science and Mathematics and my Master’s in Artificial Intelligence from KAIST, advised by Jaegul Choo. I am interested in building human-like AI, currently focusing on LLM-based agents, world models, optimization for real-time inference, and turning ML research into game experiences.
Work Experience
KRAFTON — Deep Learning Division · Research Engineer (Feb. 2025 - Present)
- PUBG Ally (official)
- Designed and implemented an on-device agent system that orchestrates STT, TTS, sLM, and BT to drive an in-game AI companion.
- Designed and implemented a custom behavior tree engine, enabling runtime LLM-driven tree editing.
- Built a pipeline that collects play data and enables sLM training and inference.
- Shipped to live service as Ally Duo, an Arcade mode in PUBG: BATTLEGROUNDS (global beta, June 2026).
-
Related Skills: UE C++, Python
NCSOFT — NC Research · Research Engineer (Mar. 2023 - Jan. 2025)
- Mobile game recommendation system
- Developed custom bandit algorithms to enhance the purchase of package items in a mobile game.
- Designed and implemented machine learning pipelines for training and inference.
- Built simulators and A/B testing tools for model evaluation and conducted on/offline tests.
- Demonstrated via focus group test that the proposed model outperformed the popularity-based model by 11.4% in purchase volume.
- Related Skills: Spark, FastAPI, Airflow, Python
- Coding assistant tools using large language models
- Developed and deployed an LLM server and copilot-like IDE extensions to enhance team productivity while preventing code leaks.
- Implemented a code interpreter feature utilizing open-source LLMs for real-time code execution and analysis within a conversational interface.
- Related Skills: React, FastAPI, vLLM, PyTorch, TypeScript, Python
Kakao Corp. — Recommendation Team · Internship (Dec. 2018 - Feb. 2019)
- Piccoma webtoon recommendation system
- Improved preprocessing of comic descriptions and tag text used in content-based filtering algorithms.
- Designed and implemented deep-learning-based contextual bandit algorithms.
- Related Skills: MongoDB, RocksDB, Elasticsearch, Apache Kafka, TensorFlow, Python
Education
- Korea Advanced Institute of Science and Technology (KAIST)
- Master’s degree in Artificial Intelligence (Feb. 2021 - Feb. 2023)
- Bachelor’s degree in Computer Science (Feb. 2017 - Feb. 2021)
- Bachelor’s degree in Mathematics (double major) (Feb. 2017 - Feb. 2021)
Awards & Certifications
- NeurIPS 2022 Scholar Award travel grant (Oct. 8, 2022)
- Advanced to ICPC World Finals Moscow Invitational Contest (Sep. 30, 2021)
- 2019 ACM-ICPC Seoul regional 3rd place (Nov. 9, 2019)
- 2018 ACM-ICPC Jakarta regional silver (Nov. 12, 2018)
- 2018 ACM-ICPC Seoul regional 5th place (Nov. 3, 2018)
- 2018 ACM-ICPC KAIST competition gold (Oct. 6, 2018)
- 2018 KAKAO code festival 4th place (Aug. 25, 2018)
- 2018 Samsung Collegiate Programming Cup 5th place (Jul. 31, 2018)
- 2017 ACM-ICPC Daejeon regional 5th place (Nov. 11, 2017)
- 2017 KAKAO code festival 5th place (Sep. 9, 2017)
For the full list of publications, see the Publications page.