AI Hardware Optimization Engineer Intern

About the Role

As an AI Hardware Optimization Intern at NeuroSpark, you will work on performance problems at the boundary between machine learning software and modern accelerator hardware. This internship is designed for students interested in GPU programming, computer architecture, compilers, parallel computing, or performance engineering. Prior industry experience is not required.

What You’ll Do

• Benchmark ML operators and model workloads on accelerators.
• Profile kernel execution, memory access, and hardware utilization.
• Help develop and test kernel-level and runtime-level optimizations.
• Experiment with tiling, operator fusion, memory reuse, and parallel execution.
• Build performance benchmarks and profiling tools.
• Compare optimization strategies across different hardware platforms.
• Work directly with engineers on production-relevant AI workloads.

Minimum Qualifications

• Currently pursuing a Bachelor’s, Master’s, or PhD in Computer Science, Computer Engineering, Electrical Engineering, or a related field.
• Programming experience in C/C++.
• Familiarity with Python.
• Coursework or project experience in systems, computer architecture, operating systems, parallel computing, or compilers.
• Interest in understanding how software performance maps to hardware behavior.

Preferred Qualifications

• CUDA or other GPU programming experience.
• Projects involving kernels, compilers, HPC, or ML systems.
• Familiarity with PyTorch, JAX, TensorFlow, or Triton.
• Experience using performance profilers.
• Understanding of GPU memory hierarchy and parallel programming concepts.

About NeuroSpark

NeuroSpark builds and operates a high-performance AI inference platform that helps enterprises run large language models faster, cheaper, and at scale. Inference infrastructure is the foundation the entire AI application layer runs on — every AI product ultimately depends on how fast, how reliably, and how affordably models can serve their users. Our vision is to make that layer so efficient that compute is never the reason a good AI product fails.