Remote AI Inference Engineer - Kernel & Performance

Lever, Inc.; Saudi Arabia; On-site; SAR 420,000 - 600,000; Full time

Benefits offered by this job

Remote-first environment

International team

Cutting-edge AI research exposure

Job summary

Lever, Inc. partners with a specialized company to search for an AI Research Engineer (Kernel & Inference Optimization) in Saudi Arabia.

The role sits at the intersection of AI research, systems engineering, and high-performance model inference, focusing on optimizing model-serving architectures for diverse hardware, including edge devices. You will tackle latency, throughput, and memory challenges, develop GPU kernels in Metal Shading Language, and apply pruning, quantization, and Flash

Qualifications

• Degree in Computer Science or related field; a PhD in NLP, ML or related discipline with publications is highly relevant.
• Proven expertise in Metal Shading Language (MSL) and writing custom compute shaders from scratch.
• Demonstrated experience with low-level kernel optimization and inference optimization on mobile or other resource-constrained devices.
• Track record of delivering measurable improvements in latency, throughput, and memory footprint for domain-specific applications.
• Deep understanding of modern model-serving architectures, inference engines, and optimization techniques for high-performance AI deployment.
• Strong experience writing GPU kernels for mobile devices.
• Practical experience developing and deploying end-to-end inference pipelines on constrained hardware.
• Strong ability to apply empirical research and systematic experimentation to solve latency, computational, and memory challenges.
• Knowledge of distributed inference techniques, including tensor parallelism, pipeline parallelism, and expert parallelism for large-scale GPU clusters.
• Deep understanding of diffusion models and Vision Transformers.
• Familiarity with pruning, quantization, Flash Attention, KV Cache optimization, and speculative decoding such as EAGLE.
• Strong analytical and problem-solving abilities and excellent English communication skills.

Responsibilities

• Design and deploy advanced model-serving architectures for high throughput and low latency.
• Develop inference pipelines across mobile and edge platforms.
• Set performance targets for latency, token generation speed, throughput, memory footprint, and reliability.
• Build controlled inference benchmarks and track latency, throughput, memory, and errors.
• Create datasets and simulation scenarios for evaluating model performance under real-world constraints.
• Identify bottlenecks across inference pipelines and implement batching, memory management, and other optimizations.
• Develop custom GPU kernels and compute shaders for mobile hardware (MSL).
• Apply pruning, quantization, Flash Attention, KV caching, and other optimization techniques.
• Design distributed inference systems using tensor/pipeline/expert parallelism for large GPU workloads.
• Collaborate with cross-functional teams to integrate inference frameworks into production and edge-device apps.
• Define evaluation methodologies and document experimental results against benchmarks.
• Monitor production performance and pursue scalability improvements.

Skills

GPU kernel optimization

Low-level optimization

Inference optimization

Distributed inference

Benchmarking

Empirical research

Cross-functional collaboration

Education

PhD in NLP or ML

Degree in Computer Science or related field

Tools

Metal Shading Language (MSL)

Flash Attention

KV Cache optimization

Job description

Lever, Inc. partners with a specialized company to search for an AI Research Engineer (Kernel & Inference Optimization) in Saudi Arabia.

The role sits at the intersection of AI research, systems engineering, and high-performance model inference, focusing on optimizing model-serving architectures for diverse hardware, including edge devices. You will tackle latency, throughput, and memory challenges, develop GPU kernels in Metal Shading Language, and apply pruning, quantization, and Flash