Nebius
Senior Applied Scientist, Efficient LLM Inference & Model Optimization
$195K–$262K
Palo Alto, California, United States +1 · Senior · Full-time · Posted Jul 22, 2026
- Location
- Palo Alto, California, United States
San Francisco Bay Area, California, United States - Education
- PhD in computer science, machine learning, ML systems, computer systems, computer architecture, electrical engineering, applied mathematics, or a closely related discipline.
- Visa sponsorship
- Applicants must be authorized to work in the country in which they apply and provide proof of employment eligibility as a condition of hire.
About the job
Research and optimize efficient LLM and VLM inference, then turn validated methods and prototypes into production capabilities.
What you'll do
- Lead inference research projects from hypotheses and experiments through prototypes and production handoff.
- Develop and evaluate optimization methods and serving architectures, measuring quality, latency, throughput, memory use, and cost.
- Collaborate with engineering and product teams, share findings through reports, papers, blogs, and open-source artifacts, and mentor colleagues.
What they ask for
- PhD in a listed or closely related discipline.
- Strong publication record or equivalent research artifacts in relevant areas.
- Strong Python and PyTorch implementation skills.
- Deep knowledge of LLMs, VLMs, transformer inference, decoding, compression, quantization, and serving trade-offs; strong experimental design and communication skills.
Nice to have
- First-author publications at listed venues; production deployment experience.
- Experience with named inference tools, post-training methods, open-source research artifacts, or distributed inference architectures.
Pay
$195,200 — $262,200 USD (base salary)
Tools
Python, PyTorch, Triton, CUDA, vLLM, SGLang, TensorRT-LLM, NVIDIA Dynamo
AI focus
Efficient LLM and VLM inference research, including quantization, distillation, speculative decoding, KV-cache optimization, routing and scheduling, distributed inference, and model/runtime co-optimization.
Model-written summary · the original posting is authoritative