This is a fully remote position, open to applicants in France, +5 more countries.

📋 Description

• Assist EMEA AI Natives customers in the deployment and optimization of extensive inference workloads on multi-node GPU clusters.

• Design efficient inference pipelines for both dense and sparse/latent MoE models, distributing workloads across thousands of GPUs.

• Enhance inference efficiency by focusing on quantization, speculative decoding, disaggregated prefill/decode, KV cache management, and WideEP for large MoE deployments.

• Collaborate with NVIDIA product teams, such as Dynamo, TensorRT-LLM, and NIXL, to boost customer success.

• Energize the AI inference developer community across EMEA through technical workshops, hackathons, and reference architectures.

• Set the technical direction for scalable, high-performance AI inference in challenging production environments.

⛳️ Requirements

• MS or PhD in Computer Science, Engineering, High-Performance Computing, or equivalent professional experience.

• Over 5 years of experience in optimizing Neural Networks inference.

• Strong understanding of transformers inference optimization, including quantization, disaggregated inference, speculative decoding, continuous batching, and KV cache optimization.

• Hands-on experience in MoE inference at scale, including expert parallelism, WideEP, all-to-all communication, routing overhead, and load balancing at scale.

• Ability to effectively engage with ML engineers, researchers, and systems architects at a profound technical level.

• Practical experience with NVIDIA Dynamo, NIXL, Grove, or emerging disaggregated inference tools.

• Knowledge of GPU memory hierarchies and high-speed interconnects, such as NVLink, InfiniBand, RDMA, and UCX.

• Contributions to advanced AI labs or large-scale AI infrastructure providers conducting inference on thousands of GPUs.

• Published work or benchmarks related to large-scale AI inference.