Kaan Aykurt
  • Home
  • CV
  • Publications
  • Teaching
  • Blog

Curriculum Vitae

Kaan Aykurt
PhD Candidate · AI-driven Networked Systems
Technical University of Munich

Summary

PhD candidate specializing in AI for Networking and Networking for AI. I develop agentic frameworks and benchmarks for autonomous network operations using Large Language Models (LLMs), and investigate the networking requirements of large-scale AI/ML workloads.

Experience

Chair of Communication Networks, Technical University of Munich
Munich, Germany
Doctoral Researcher
2022 – present
  • Measuring and modeling how close TCP gets to RoCE for LLM serving and fine-tuning on Ethernet GPU clusters, showing it stays within 4% of RoCE in pipeline-parallel serving and predicting capacity within 7.2% for a 32B model.
  • Distilling small language models to assess their improvement potential for autonomous network management, nearly tripling their success rate on operational networking tasks.
  • Built RouteMind, an LLM agent framework that generates BGP configurations and negotiates routing policy between autonomous systems in emulated networks.
  • Designed NetLLMBench, an open benchmark that checks LLM-generated network configurations in emulation.
  • Analyzed on a hardware testbed how TCP and QUIC behave when data center networks are reconfigured at runtime, and predicted the resulting slowdowns with an ML model that explains 88% of their variance.
  • Built and operated a Kubernetes experimental cluster and developed HyPA, a pod autoscaler that adapts to changing request patterns.
  • Redesigned the Programmable Communication Networks Lab around containerlab, infrastructure-as-code automation and a class-wide mini-internet project; supervised seven student theses and research projects.
Working Student
2020 – 2022
  • Characterized the network traffic of distributed ML training (PyTorch, MPI) and compared RoCE with TCP.
  • Applied genetic algorithms to adversarial benchmarking of data center networks and Kubernetes clusters, generating worst-case workloads that maximize flow completion times and degrade service performance.
IBM
Istanbul, Türkiye
IT Specialist
2018 – 2019
  • Worked on the server monitoring stack of a top-three Turkish bank with IBM Tivoli Netcool and Grafana.

Technical Skills

AI & ML Python, PyTorch, Hugging Face, vLLM, Distributed Training, LoRA Fine-Tuning, LLM Agents, MCP
Networking TCP/IP, BGP, OSPF, RDMA, RoCE, InfiniBand, NCCL, P4, FRR, Containerlab
Infrastructure Linux, Docker, Kubernetes, Helm, Git, GitLab CI, Prometheus, Grafana
Modeling & Analysis Performance Modeling, Measurement Studies, Benchmarking
Languages English (fluent), German (intermediate), Turkish (native)

Education

Technical University of Munich, Germany
Expected 2027
Ph.D. in Computer Engineering (Advisor: Prof. Wolfgang Kellerer)
Technical University of Munich, Germany
2022
M.Sc. in Communications Engineering
Koç University, Istanbul, Türkiye
2019
B.Sc. in Electrical & Electronics Engineering and B.A. in Economics (Double Major)
 

© 2026 Kaan Aykurt. Licensed under CC BY 4.0.