About the role

VESSL AI is a GPU Cloud company providing GPU infrastructure for AI companies. With explosively growing demand for AI compute, we help AI companies reliably secure the GPU computing power they need, exactly when they need it, through GPUaaS (GPU-as-a-Service). Today, our products are used by a wide range of companies and startups in Korea, including Upstage, SqueezeBits, and Holiday Robotics, as well as leading academic and frontier labs in the US such as UC Berkeley, Stanford, Subquadratic, and Nuance Labs. As of 2026, we have grown more than 20x year-over-year, establishing ourselves as one of the fastest-growing companies in the global AI infrastructure market.

Unlike other GPU Clouds, VESSL AI doesn't stop at simply supplying GPUs. We develop a deep understanding of our customers' AI workloads and deliver optimal infrastructure solutions built on that understanding. To do this, our R&D team validates and benchmarks a broad range of models, and directly builds Post-Training and Inference optimized for agentic workloads. The team's research currently centers on two core pillars. The first is RL based Post-Training for Agentic AI, research that uses reinforcement learning to strengthen the reasoning, tool-use, and coding abilities demanded by real-world agentic workloads. The second is distributed cache optimization and cutting-edge accelerator kernel optimization for large-scale Agentic AI model serving, pushing inference efficiency to its limits on the latest GPUs, from H200 to B200, B300, and GB300.

As a core member of this R&D team, the Senior AI Research Engineer owns LLM Inference optimization and Post-Training directly, taking the lead in defining and solving problems within whichever of these two pillars best matches their expertise. You'll conduct large-scale, frontier research on top of abundant GPU resources, and translate the results directly into VESSL's products and customer experience.

Key Responsibilities

1. LLM serving/inference optimization research: Improve the inference efficiency of framework-based engines such as vLLM and SGLang in large-scale GPU cluster environments, focusing on Multi-token Prediction, Speculative Decoding, Lossless Compression, and Kernel Optimization (CUDA, Triton, etc.)
2. LLM Post-Training research: Design multi-node, large-scale post-training pipelines using verl, vime, and slime, and directly implement and optimize Reinforcement Learning and On/Off-policy Distillation across the Reasoning/Math/Code domains
3. Research-to-product: Reflect research outcomes into VESSL's Inference/Post-Training product features and real customer workloads through benchmarking and model validation
4. R&D direction-setting and team leadership: Drive discussions on research priorities and roadmaps, and strengthen the team's research capabilities through code/design reviews, technical documentation, and mentoring of junior engineers
5. Tracking and sharing the latest research: Continuously review the latest papers and open-source projects in Agentic AI and LLM serving/training, and share insights through internal study groups and seminars

Qualifications

1. PhD in Computer Science, Electrical Engineering, AI/ML, or a related field; or an MS plus 4+ years of hands-on/research experience
2. Experience driving meaningful inference performance gains in one or more of: Speculative Decoding, Lossless Compression, or Kernel Optimization (CUDA, Triton, etc.)
3. Experience improving model performance through fine-tuning or post-training of Large Language Models (LLMs)
4. Ability to quickly read the latest papers and implement/validate their core ideas in code
5. Experience taking the lead on research roadmaps or technical decision-making in team projects or research labs
6. Business-level English communication skills

Helpful experience (not required)

1. Publications at international ML/systems venues such as NeurIPS, ICML, or MLSys
2. Contributions to open-source projects such as vLLM, SGLang, verl, or slime
3. Deep understanding of RL algorithms (GRPO, PPO, etc.) and the latest inference optimization techniques such as FlashAttention, Quantization (MXFP4/NVFP4/W4AFP8), and Continuous Batching
4. Experience contributing to team growth through researcher/engineer hiring, onboarding, and mentoring
5. Understanding of multi-node GPU cluster operations (Slurm, Kubernetes, etc.)
6. Comfortable reaching across SWE/infrastructure/product boundaries and getting hands-on whenever it's needed to solve a problem, rather than staying confined to a fixed scope
7. Comfortable digging in independently until you find the answer, even when facing ambiguous, open-ended problems

Joinning Process

Document Screening → Take-Home Assignment → Technical Interview → Resume/Culture Interview → CEO Interview

1. The above outlines VESSL AI Korea's standard hiring process for experienced professionals; the process may vary depending on the role.
2. Please be sure to submit your application (detailed work experience) along with your portfolio (or a link to your Git repository). There is no required format.
3. The Technical Interview focuses on technical discussion — covering topics such as system design and implementation approaches — to assess problem-solving ability, and typically takes up to 2 hours.
4. The Resume/Culture Interview focuses on relevant experience to evaluate technical competency and cultural fit, and includes the hiring manager and team members. Each session takes approximately 1 hour.
5. For experienced-level candidates, a Reference Check will be conducted after the final interview stage.
6. If any false information is found in your resume or submitted documents, your offer may be rescinded even after acceptance has been announced.

Employment Type

1. Full-time (with a 3-month probationary period)
2. Final confirmation of employment will be determined based on a performance evaluation following the 3-month probationary feedback period.