Senior AI Scientist: LLM/VLM Inference & Optimization

Nebius Group; Netherlands; Remote; EUR 120,000 - 180,000; Full time

Benefits offered by this job

Competitive compensation

Career growth

Flexibility and ownership

Collaborative culture

Impactful AI projects

International environment

Job summary

Nebius is hiring researchers to tackle frontier LLM and VLM inference bottlenecks. You will own focused research, write robust code, and produce prototypes, publications, and artifacts for engineers to build production-ready components.

Your work spans quantization, distillation, speculative decoding, KV-cache optimization, and model/runtime co-optimization, evaluating trade-offs in quality, latency, throughput, memory, and cost per token with strong collaboration across teams.

Qualifications

• PhD in CS or related field with strong publication record.
• Strong Python and PyTorch implementation skills.
• Deep knowledge of LLMs, VLMs, and decoder inference.
• Experience with model optimization and quantization.
• Excellent experimental design and communication.

Responsibilities

• Lead research projects in efficient LLM and VLM inference from hypotheses to production handoff.
• Develop methods across quantization, distillation, KV-cache reuse, and co-optimization.
• Build prototypes using PyTorch, Triton, CUDA, or inference-serving frameworks.
• Study routing, scheduling, and cache strategies to optimize latency and cost.
• Investigate multi-node inference and WideEP for expert parallelism.
• Conduct rigorous evaluations of quality, latency, throughput, and memory.
• Collaborate with MLEs, backend teams, and products to drive impact.
• Share results via reports, blogs, papers, and open-source artifacts.

Skills

Python

PyTorch

LLMs

VLMs

quantization

distillation

model compression

experimental design

communication

Education

PhD in computer science

Tools

PyTorch internals

Triton

CUDA

TensorRT-LLM

FlashAttention

Job description

Nebius is hiring researchers to tackle frontier LLM and VLM inference bottlenecks. You will own focused research, write robust code, and produce prototypes, publications, and artifacts for engineers to build production-ready components.

Your work spans quantization, distillation, speculative decoding, KV-cache optimization, and model/runtime co-optimization, evaluating trade-offs in quality, latency, throughput, memory, and cost per token with strong collaboration across teams.