Deep Learning Div. Research Scientist/Engineer (2+ years / Intern)

Introducing Our Team (Project)

Deep Learning Division Vision

KRAFTON's Deep Learning Division collaborates with various internal and external fields to provide AI solutions for diverse problems and develops its own services through proprietary deep learning technology research. The direction is broadly twofold:

• Increasing game production efficiency and universality through the development of deep learning technology that can assist in game production.
• Developing CPC/LLM Agents that can play games with users.

Proprietary AI Foundation Model Project R&D

As the cornerstone of the Deep Learning Division's second direction, "Developing CPC/LLM Agents that can play games with users," the proprietary AI Foundation Model Project is researching and developing a Large Multi-modal Model capable of simultaneously generating dialogue and and actions. Specifically, the short-term goal is to implement natural Full-Duplex interaction similar to real human conversation, while also performing accurate actions in real-time according to user requirements. This involves (1) designing the model architecture and (2) building data. Some of the developed models and data are planned to be open-sourced and will contribute to PUBG CPC services through continuous advancement.

Culture Fit

Members of the Deep Learning Division can interact and collaborate with team members from various fields through diverse projects, offering creative ideas for various problems. An atmosphere that encourages free expression of opinions, regardless of age or position, is fostered. The team is composed of individuals from diverse cultural backgrounds, and methods to overcome language barriers, such as interpretation and translation, are actively supported for lively communication.

Introducing the Missions You Will Undertake with Our Team

Modeling

• Research on expanding 32B-scale Dense/MoE LLM models to Speech (Dialogue) modality.
• Exploration and design of model architectures based on recently published Multi-modal LLM research.
• Optimization of model structure for real-time serving.
• Construction and management of datasets to maximize model performance.

Pre/Post-training

• Exploration and design of post-training pipelines based on recently published technical reports.
• Exploration and optimization of SFT/RL training frameworks for Dense/MoE models for large-scale training.
• Construction and management of benchmarks for model evaluation.

We want to grow with someone who has these experiences! (Required Qualifications)

• Ability to lead research, quickly understand deep learning papers, and design experiments.
• Ability to quickly adapt to new domains.
• Excellent work communication skills.
• No disqualification for overseas business trips.

If you have these experiences, you are the one we are looking for! (Preferred Qualifications)

• Experience in LLM/(Speech) LMM training and serving.
• Master's/Ph.D. degree or equivalent research/practical experience in deep learning related fields.
• Experience writing papers for top-tier AI/ML conferences (NeurIPS, ICLR, ICML, etc.) or top-tier journals in related fields.