Neural Network Performance Engineer

About the Company

Humanoid is on a mission to build the world’s most capable, commercially-scalable, and safe humanoid robots. Their HMND-01 Alpha platform is already running in real industrial pilots.

Responsibilities

• Analyze performance bottlenecks of model architectures and propose improvements
• Optimize model execution on new hardware such as NVIDIA Thor
• Implement custom kernels to reduce memory throughput requirements
• Quantize models with minimal loss of quality
• Suggest and implement architectural changes to improve performance without sacrificing capabilities

Requirements

• 3+ years building deep-learning systems in industry or research
• 1+ years experience in neural network inference performance (bottleneck analysis, custom kernels, quantization, compilers)
• Excellent understanding of GPU architecture
• Strong proficiency in Python and PyTorch or JAX
• Ability to profile, debug numerics, and write maintainable research code
• Clear documentation of experiments and communication of trade-offs

Preferred Qualifications

• Experience in robotics or autonomous driving
• Open source code showcasing inference performance improvements
• Publications at ICLR, ICML, NeurIPS, or equivalent contributions
• Familiarity with vision-language (VLM) or vision-language-action (VLA) models

Benefits

• Competitive equity and stock options
• 30+ paid days off including annual leave and bank holidays
• Private healthcare (virtual and in-person)
• Pension scheme with 8% total contribution
• Free daily breakfast, catered lunch, and snacks in-office