Senior Machine Learning Engineer, LLM Inference Optimization
About this role
Evaluating serving configurations, diagnosing performance bottlenecks, and optimizing inference for real-world workloads form the daily routine in this position. The team builds fast, reliable, and cost-efficient inference services for frontier models at Nebius Token Factory, scaling a full-stack AI cloud platform for developers and enterprises globally.
Ownership of model and endpoint optimization spans from model artifacts through production deployment, involving deep work on inference engines, serving architecture, and benchmarking. Success in this role requires a hands-on engineer capable of driving tangible improvements in latency, throughput, memory efficiency, and GPU utilization while preserving model quality.
The organization maintains a global R&D footprint and values bold thinking, trust, and real ownership. Ideal candidates thrive in fast-moving environments and bring deep technical expertise in transformer architectures and production machine learning systems.
What you will do
- Manage end-to-end optimization projects for specific model families, customer endpoints, and serving backends.
- Run engine comparisons and recommend practical serving configurations tailored to specific production workloads.
- Debug model quality and performance regressions during software rollouts.
- Deploy, configure, benchmark, and extend inference engines such as vLLM, SGLang, TensorRT-LLM, and Triton Inference Server.
- Build and productionize model-compression workflows involving quantization, quantization-aware training, and distillation.
- Partner with GPU kernel and platform engineers to diagnose bottlenecks across model code, runtimes, and cluster layers.
- Author design documents, performance reports, rollout plans, and customer-facing technical explanations.
What you need
- Demonstrate strong software engineering skills in Python and PyTorch.
- Bring hands-on experience deploying or optimizing large language model, vision-language model, or transformer inference systems.
- Possess practical knowledge of at least one modern inference stack including vLLM, SGLang, TensorRT-LLM, or Triton Inference Server.
- Show a strong understanding of transformer inference bottlenecks such as KV cache, attention, memory bandwidth, and batching.
- Apply quantitative reasoning to evaluate tradeoffs involving latency, throughput, quality, utilization, and cost.
- Communicate effectively and collaborate across research, kernel, infrastructure, product, and customer teams.
Nice to have
- Experience with quantization-aware training, post-training quantization, FP8, INT8, INT4, AWQ, GPTQ, or SmoothQuant techniques.
- Experience with distillation, speculative decoding, EAGLE, Medusa, or other inference acceleration methods.
- Experience handling agentic workloads, tool calling, structured outputs, streaming APIs, and high concurrency.
- Familiarity with CUDA or Triton outside of primary kernel-engineering responsibilities.
- Open-source contributions to projects like vLLM, SGLang, TensorRT-LLM, FlashInfer, LMCache, PyTorch, or Ray.
Benefits
- Competitive compensation
- Career growth and learning opportunities
- Flexibility and ownership
- Collaborative and innovative culture
- Opportunity to work on impactful AI projects
- International environment and talented teams
Skills and tech stack
Questions candidates ask
What is the work location policy for this position?
This position is offered as remote within Europe. Nebius operates with a global footprint and R&D hubs across Europe, the UK, North America, and Israel.
What technologies and inference engines will I be working with?
The tech stack includes Python, PyTorch, and modern inference engines such as vLLM, SGLang, TensorRT-LLM, Triton Inference Server, and NVIDIA Dynamo. Engineers also work with model-compression workflows, quantization techniques, and hardware layers alongside GPU kernel engineers.
What does day-to-day collaboration look like on the team?
You will work closely with kernel and platform engineers to evaluate serving configurations, resolve performance regressions, and optimize inference for real-world workloads. The role also involves partnering with research, infrastructure, product, and customer teams.
What kind of environment can I expect at Nebius?
The company culture is fast-moving with bold thinking, constant growth, meaningful impact, trust, and real ownership. Teams operate in an international environment focused on shaping the future of AI.