Large-scale AI Model Training and Optimization Engineer

Responsibilities

• Design and optimize GPU computing kernels for large-scale foundation model pre-training and inference
• Develop and design distributed learning and Reinforcement Learning (RL) based training pipelines
• Research and implement asynchronous learning architectures, including rollout–trainer separation and actor–learner scheduling
• Optimize high-speed inference using techniques such as quantization, kernel fusion, KV cache, and paged attention
• Improve training and serving throughput and latency by optimizing computation-communication overlap and memory/bandwidth efficiency
• Research and apply the latest learning algorithms to production-level training systems

Requirements

• 3+ years of professional experience
• PhD in AI, Computer Science, Electrical Engineering, or a related field (or a Master's degree with 3+ years of relevant experience)
• Proven experience in LLM or large-scale deep learning model training and inference development
• Proficiency in Python, PyTorch, and Linux environments
• Experience developing in distributed learning environments

Preferred Qualifications

• Experience with GPU kernel-level optimization using CUDA, Triton, or CUTLASS
• Experience developing large-scale distributed training (e.g., Megatron) and RL infrastructure (RLHF, RLAIF)
• Experience with Ray or designing asynchronous/event-driven distributed systems
• Experience contributing to or deeply utilizing inference engines like vLLM, SGLang, or TensorRT
• Direct experience performing LLM pre-training or post-training (fine-tuning, RL)
• Publications in top-tier AI conferences (NeurIPS, ICML, ICLR, MLSys, etc.)
• Contributions to open-source projects

About the Company

SK telecom is a leading technology company dedicated to creating a happier world through advanced digital innovation and AI-driven solutions.