ABOUT THE OPPORTUNITY

Help a leading AI lab improve how its models reason about the systems that run large language models.

This opportunity is for engineers with practical experience in ML infrastructure, accelerator performance, distributed workloads or high-throughput inference serving. You will turn that expertise into challenging technical tasks, reference solutions and evaluations that help improve frontier AI systems.

HumanitApp is sharing this opportunity as an independent referral curator. Depending on your location, the engagement is arranged through Cincinnatus LLC or one of its local employer-of-record partners.

WHAT YOU WILL DO

- Design realistic tasks and write reference solutions involving accelerator kernels, performance profiling, distributed debugging and LLM inference serving.
- Create challenging examples that reveal weaknesses in AI-generated ML systems solutions.
- Evaluate technical tasks and solutions, explaining your findings in clear written feedback.
- Develop evaluation criteria for kernel optimization, profiler interpretation, distributed systems reasoning, throughput, latency and memory trade-offs.
- Collaborate with researchers and other specialists to improve the accuracy and consistency of AI training data.





REQUIRED EXPERIENCE

- At least two years of professional experience in ML systems, ML infrastructure, model serving or accelerator performance engineering.
- Hands-on expertise in at least one of these areas: custom kernel development or optimization; performance profiling and trace analysis; debugging distributed or accelerator-bound workloads; or serving LLMs at scale.
- Production experience with JAX and/or PyTorch.
- Experience with modern accelerators, such as TPUs, NVIDIA H100/B200, AWS Trainium or AMD MI300, and an understanding of throughput, latency and memory trade-offs.
- Demonstrable skilled progression and strong written communication.
- Reliable availability for 40 hours per week during weekdays, without conflicting or concurrent engagements.
- JAX, XLA or TPU expertise.
- Kernel tools such as Pallas, Triton, CUDA or CUTLASS.
- Profiling tools such as JAX/XLA profiler, Kineto, torch.profiler or Nsight.
- LLM serving systems such as vLLM, SGLang, TensorRT-LLM or JetStream.
- Framework or compiler work involving custom operators, sharding, distributed training, XLA/StableHLO, torch.compile or MLIR.

📌 MLOps Engineer, LLM Performance (India)
🏢 HumaniTapp
📍 India