Reinforcement Learning for Kubernetes Autoscaling
Nokia engineering work on DDQN reinforcement learning, time-series prediction and configurable training pipelines for Kubernetes pod autoscaling.
What did the autoscaling work address?
At Nokia, I worked on models for Kubernetes pod autoscaling, including a DDQN reinforcement learning model, PyTorch deep learning models and tuned open-source time-series predictors. The work connected machine learning experiments with cluster infrastructure.
My role
During my Nokia roles (2022–2025), I designed, created and deployed models, built configurable training pipelines, and configured CI/CD and lab deployments.
Training and deployment architecture
Training pipelines and model registries ran in a remote GPU-powered lab. Jenkins pipelines supported CI/CD, while Helm charts configured deployments in OpenShift labs. The work involved both model development and the infrastructure needed to train and deploy those models.
Technologies
Reinforcement learning, DDQN, Python, PyTorch, Kubernetes, Jenkins, Helm and OpenShift.
Scope of available evidence
This description is based on my professional experience recorded in the CV. No public code, workload traces or comparative benchmark is linked, so I do not claim a quantified resource saving or advantage over other autoscalers.