Requirements

PhD in computer science, machine learning, ML systems, computer systems, computer architecture, electrical engineering, applied mathematics, or a closely related discipline.
Strong publication record or equivalent research artifacts in ML, ML systems, efficient inference, model compression, quantization, distillation, serving systems, or related areas.
Strong Python and PyTorch implementation skills, capable of turning ideas into experiments and working prototypes.
Deep knowledge of LLMs, VLMs, transformer inference, decoding algorithms, model compression, quantization, and production-serving trade-offs.
Strong experimental design skills covering ablations, baselines, metrics, statistical reasoning, and failure analysis.
Excellent written and verbal communication skills.

Nice to haves

First-author publications at top-tier conferences such as NeurIPS, ICML, ICLR, MLSys, ACL, EMNLP, ASPLOS, OSDI, SOSP, ISCA, or HPCA.
Experience deploying ML models or inference optimizations in production environments.
Familiarity with vLLM, SGLang, TensorRT-LLM, NVIDIA Dynamo, FlashAttention, FlashInfer, Triton, CUDA, or PyTorch internals.
Experience with post-training, SFT, DPO, RLHF, RLAIF, preference optimization, or synthetic data generation to improve inference quality or efficiency.
Open-source research artifacts, popular benchmarks, technical blogs, or invited talks showing contributions to efficient AI systems.
Research or implementation experience in distributed inference architectures such as LLM request routing, prefill–decode disaggregation, multi-node serving, or MoE expert parallelism including WideEP.

Lead research projects to improve efficient LLM and VLM inference, handling hypothesis formation, experimentation, ablations, prototyping, and production handoffs.
Develop and evaluate quantization, quantization-aware training, distillation, speculative decoding, KV-cache reuse/compression, long-context inference, MoE routing, and model/runtime co-optimization methods.
Build prototypes with PyTorch, Triton, CUDA-adjacent tools, or inference-serving frameworks, collaborating with MLEs and platform engineers to transition prototypes into production.
Research LLM request routing and scheduling including cache-aware load balancing and prefill–decode disaggregation to optimize latency, throughput, and cost.
Investigate multi-node inference for dense and mixture-of-experts models, focusing on parallelism, expert placement, load balancing, and computation/communication trade-offs to guide architectural choices.
Conduct rigorous evaluations on quality, latency, throughput, numerical stability, memory footprint, tail latency, and cost per token under consistent workload and GPU conditions.
Collaborate with MLE, GPU kernel, backend, product, and customer teams to prioritize impactful production research.
Share findings via reports, technical blogs, papers, and open-source work; mentor engineers and scientists on experimental design and the trade-offs involved.