Senior Applied Scientist, Efficient LLM Inference & Model Optimization
Core
Research and optimize efficient inference for frontier LLMs and VLMs, translating research findings into production capabilities for the Nebius AI cloud platform.
Role type
Senior Applied Scientist (Efficient LLM Inference & Model Optimization)
Builds
Production-ready inference engines, optimized serving architectures, and distributed inference systems for LLMs/VLMs.
Domain
Artificial Intelligence / Machine Learning Systems / High-Performance Computing
Deliverable
production ML models
Required skills
LLM/VLM inference, transformer decoding algorithms, model compression, quantization, distillation, speculative decoding, KV-cache optimization, distributed inference architecture, experimental design, PyTorch, CUDA, Triton
Preferred skills
vLLM, SGLang, TensorRT-LLM, FlashAttention, FlashInfer, MoE routing, PDD, WideEP, post-training optimization (SFT/DPO/RLHF)
Technologies
PyTorch, Triton, CUDA, vLLM, SGLang, TensorRT-LLM, FlashAttention, FlashInfer
Responsibilities
Lead research projects in efficient LLM/VLM inference from hypothesis to production handoff; Develop and evaluate methods spanning quantization, distillation, and speculative decoding; Build prototypes using CUDA-adjacent tooling and collaborate with MLEs to productionize; Research LLM request routing, scheduling, and prefill–decode disaggregation; Investigate multi-node inference and MoE expert parallelism; Develop rigorous evaluations covering latency, throughput, and cost per token.
Seniority
Senior, hands-on IC with research leadership
