Victor Ruehle

À propos

Victor Ruehle is a Partner Research Manager at Microsoft, where he leads the Efficient AI research pillar within M365 Research. His team reimagines AI efficiency across the full stack, from GPU kernels and hardware-aware model execution through large-scale inference systems to context engineering and agents. Its goal is to make Copilot-scale AI faster, more reliable, more sustainable, and ready to scale.

The team’s research is organized around three areas:

• Kernel-level innovation and hardware-aware modeling
• System-level innovation for inference at scale
• Context engineering and agents

Across these areas, Victor’s team develops end-to-end optimizations spanning models, scheduling, batching, routing, resource allocation, context, and memory. The aim is to unlock multiplicative improvements in cost, latency, and reliability by jointly optimizing across layers rather than treating each layer in isolation.

Recent research spans heterogeneous LLM serving, workload-aware routing, hardware-aware attention, prompt compression, adaptive retrieval and reasoning, and memory for long-running agents. This work has been published in Nature and at leading venues including ICML, ICLR, ACL, MLSys, ACM SIGMETRICS, The Web Conference, and AAMAS.

Before focusing on Efficient AI, Victor worked on privacy-preserving machine learning and AI for cloud operations. He earned his PhD in physics from the Max Planck Institute for Polymer Research and previously worked as a postdoctoral researcher at the University of Cambridge and as a physical-property specialist in the oil and gas industry.