Neural Network Performance Engineer

Here at Humanoid, we believe in a future where robots amplify human potential. That’s why we’ve set out on a mission to build the world’s most capable, commercially-scalable, and safe humanoid robots. We’re bringing that mission to life with HMND‑01 - our rapidly developed humanoid platform being deployed in real industrial environments - and we’re growing the team to take it even further.

We're hiring a Neural Network Performance Engineer to join our VLA team based in London. In this role, you will work on all aspects of running capable neural-network based control policies at a high rate with minimal latency, both on cloud hardware and onboard. Your work will be critical to delivering smooth robot motions while reacting to environment changes as quickly as possible.

• →

  Analyze performance bottlenecks of a particular model architecture and come up with potential improvements.

• →

  Make the model run on a new hardware (e.g. NVIDIA Thor) efficiently.

• →

  Implement custom kernels to reduce memory throughput requirements where it matters.

• →

  Quantize a model with minimal loss of quality.

• →

  Suggest and implement changes of model architecture that will enable better performance characteristics without sacrificing model capabilities.

• 3+ years building deep‑learning systems (industry or research) with shipped models or published artifacts to show for it.

• 1+ years experience working on performance of neural network inference (analyzing bottlenecks, writing custom kernels, quantizing models, fighting deep learning compilers).

• Excellent understanding of GPU architecture and why some models run faster than others.

• Strong Python + PyTorch/JAX; you can profile, debug numerics, and write maintainable research code.

• You document experiments clearly and communicate trade‑offs crisply.

  Nice to have:

• Robotics or autonomous driving experience.

• Open source code showcasing your ability to improve inference performance.

• Publications at ICLR/ICML/NeurIPS or equivalent open‑source contributions.

• Familiarity with vision-language (VLM) or vision-language-action (VLA) models.