CUDA / AI Engineer

MWDN • Madrid, MD

16h

About The Position

MWDN is seeking a highly technical engineer or researcher with deep expertise in GPU computing, CUDA, and LLM inference optimization. The role focuses on challenging problems at the intersection of AI systems and high-performance computing, for someone interested in understanding systems from GPU architecture and low-level kernels through LLM inference engines and performance bottlenecks.

The engineer will investigate and optimize LLM inference performance on modern GPUs, work with inference and serving frameworks, and analyze GPU execution, memory access patterns, kernel performance, and inference bottlenecks. The role also involves developing, optimizing, debugging, or reverse-engineering CUDA kernels and low-level GPU operations; exploring optimizations for matrix multiplication, attention mechanisms, and neural network layers; and benchmarking performance to improve latency, throughput, and GPU utilization.

Requirements

• Strong hands-on experience with CUDA and GPU programming.
• Deep understanding of GPU architecture, memory hierarchy, parallel execution, and performance optimization.
• Experience with LLM inference, AI systems, GPU kernels, or high-performance computing.
• Familiarity with one or more of vLLM, SGLang, TensorRT-LLM, Triton, or FlashAttention.
• Experience debugging complex performance issues and investigating low-level system behavior.
• Strong problem-solving skills.
• Ability to communicate complex technical concepts in English.

Nice To Haves

• PhD or research experience in Computer Science, Machine Learning, Computer Architecture, or a related field.
• Publications, research projects, or meaningful contributions to open-source GPU/LLM infrastructure.
• Experience with PyTorch internals, distributed inference, quantization, KV cache optimization, or custom GPU kernels.
• Contributions to projects such as vLLM, SGLang, Triton, FlashAttention, or related repositories.

Responsibilities

• Investigate and optimize LLM inference performance on modern GPUs.
• Work with inference and serving frameworks such as vLLM, SGLang, TensorRT-LLM, or similar systems.
• Analyze GPU execution, memory access patterns, kernel performance, and inference bottlenecks.
• Develop, optimize, debug, or reverse-engineer CUDA kernels and low-level GPU operations.
• Explore optimization techniques for matrix multiplication, attention mechanisms, and neural network layers.
• Benchmark performance and identify opportunities to improve latency, throughput, and GPU utilization.
• Tackle technically challenging problems through analytical thinking and hands-on experimentation.

Benefits

• Flexible working hours.
• 29 days of PTO (18 working days per year plus all national holidays).
• 10 paid recovery days.
• Full financial and legal support for independent contractors.
• Free English classes with native speakers or Ukrainian teachers.
• Dedicated HR support.
• Professional growth opportunities and team-building events.
• Career support to help find new opportunities if a project is not the right fit.

What This Job Offers

• Job Type: Full-time
• Education Level: No Education Listed