ML-Driven Kubernetes NetworkPolicy Generation
Master thesis · eBPF · Cilium · Hubble · Kubernetes
Designed a data-driven system that learns baseline pod-to-pod communication from eBPF flow data (Cilium/Hubble) and automatically synthesizes Kubernetes NetworkPolicy objects to enforce least-privilege segmentation in a single-cluster environment. Implemented and evaluated the approach on real-world microservice benchmarks (Train Ticket, Sock Shop), using a supervised ML classifier over curated “known-good” traffic and avoiding iterative retraining to reduce the risk of adversarial poisoning.