About the AI Research Division
KRAFTON AI Research Division is building a next-generation AI-based game production paradigm centered on large language models (LLMs). We directly perform the entire process from model design, data construction, large-scale training, optimization, evaluation, deployment, and operation, aiming to optimize game production efficiency and create new play experiences. Our direction broadly covers two areas:
- Increasing game production efficiency and universality through deep learning technology development
- Developing Co-Playable Characters that can play games with users
Foundation Model Engineering Team Introduction
The Foundation Model Engineering team implements the research direction of the AI Research Division into actual large-scale models and systems. Research Engineers transform research ideas and paper-level hypotheses into actual trainable models, reproducible experiments, scalable pipelines, and stable evaluation/deployment systems.
- Handles distributed training, memory optimization, communication optimization, checkpoint management, and experiment automation technologies for stable training of models with billions to tens of billions of parameters.
- Specifies the direction proposed by researchers into executable experimental units, analyzes results, and creates the basis for proposing the next research direction.
- Establishes an end-to-end Foundation Model development system from model training to inference, evaluation, and deployment.
Culture Fit
The AI Research Division is an organization where members with diverse expertise define and solve problems together. We collaborate with teammates from various fields, including researchers and engineers, and encourage opinion expression based on technical evidence regardless of position or years of experience. Developing large-scale Foundation Models involves inseparable issues of modeling, data, systems, and evaluation. Therefore, rather than staying in one area, we value the attitude of self-tracking where the bottleneck of the problem lies and delving into the necessary areas.
Your Mission
As a Research Engineer, you will perform the following tasks:
- Design and implement end-to-end pipelines for training, evaluation, inference, and deployment of large-scale LLM and Multi-modal Foundation Models.
- Implement learning algorithms, data construction methods, optimization strategies, and evaluation methodologies into actual code and experimental systems to improve model performance and stability.
- Lead the entire process of experiment design, implementation, execution, analysis, and documentation by concretizing research directions and hypotheses into verifiable experimental units.
- Apply and improve model parallelism, data parallelism, pipeline parallelism, communication optimization, memory optimization, and checkpoint strategies in large-scale distributed learning environments.
- Integrate new model architectures, learning techniques, and optimization techniques into existing learning codebases and experimental pipelines, and verify them experimentally.
- Build a reproducible and repeatable experimental environment to create an experimental system that can clearly analyze the causes of model improvements.
- Analyze and solve problems such as performance degradation, instability, convergence failure, distributed system errors, data bottlenecks, and evaluation discrepancies that occur during model training.
- Quantitatively analyze model performance, learning efficiency, system efficiency, and evaluation results, and share them in a format that researchers and engineers can use for decision-making.
Required Experience
- Bachelor's degree or higher in AI, Computer Science, Statistics, Electrical and Electronic Engineering, or a related field, or equivalent practical experience.
- Research and development experience in machine learning, deep learning, and Foundation Models.
- Software engineering capabilities to understand and modify models, learning loops, data loaders, evaluation pipelines, and experimental settings in large-scale codebases.
- Ability to quantitatively analyze experimental results, hypothesize the causes of performance changes, and connect them to subsequent experiments.
- Excellent communication skills to clearly document and share complex implementation details, experimental settings, and result analysis.
- Ability to structure and solve problems in collaboration with researchers, engineers, and product organizations.
Preferred Experience
- Experience in training, evaluating, and inferring large-scale LLM or Multi-modal Foundation Models.
- Experience in handling systemic problems required for large-scale model development, such as distributed learning, learning optimization, and model parallelism.
- Experience in analyzing and solving complex problems in the Foundation Model development process, such as learning instability, performance bottlenecks, evaluation discrepancies, and data quality issues.
- Experience in designing or improving ML pipelines from model training to evaluation, inference, and deployment.
- Understanding or experience in cutting-edge research topics such as LLM agents, reasoning, tool use, reinforcement learning, and multi-modal learning.
- Strong ability to implement research ideas into actual working code and reproducible experiments.