GAUSS
Large language model (LLM) inference has become an operational workload that fleets must provision, monitor, and bill against latency service-level objectives (SLOs). A single throughput number cannot answer the questions operators actually face, because generative…
Capillary Kernels
Tree-structured prefix sharing is now standard in large language model (LLM) serving: a radix-tree KV cache lets multiple requests reuse common context prefixes and reduces memory capacity pressure. Yet the attention kernels used with these…
Principal Applied Scientist
Copilot Connectors team works across the stack, from crawl systems, data modelling, information extraction, indexing, query and retrieval, integrations into different hardness system, writing skills and plug-ins and hill climbing by building Eval systems. As…
Senior/Principal Optical Packaging Engineer
At Microsoft Research, we are shaping the next generation of AI infrastructure through long-horizon programmes spanning devices, systems, software and hardware. As a Principal Optical Packaging Engineer, you will own the package architecture and assembly…