keda

KEDA External gRPC Scaler for GPU workloads - native NVML metrics via DaemonSet, no Prometheus required

Multi-cloud infrastructure deployment for the HyperI Data Fusion Engine (OpenTofu, Helm, Argo CD).

Kubernetes autoscaling experiment on AKS: HPA vs KEDA and CPU vs request-rate scaling, 180 open-loop runs with raw per-request data, analysis and the thesis research log.

OpenAI-compatible AI execution plane on Kubernetes with MCP, vLLM, and policy.

DOKS 환경에서 ML 기반 예측 오토스케일러(LSTM/GRU/Ensemble)와 Kubernetes HPA를 4개 트래픽 archetype에 걸쳐 비교 평가. KEDA 연동, 84회 실험, 리드 타임 측정 도구 자체 개발. KSCI 논문.

Archived — K3s cluster documentation: Wazuh security, KEDA autoscaling, and monitoring stack.

Cut GPU and AI workload costs on AWS: Spot and right-sized GPU nodes, scale-to-zero with Karpenter and KEDA on EKS, and cost-per-job dashboards in Prometheus and Grafana.

KEDA is a Kubernetes-based Event Driven Autoscaling component. It provides event driven scale for any container running in Kubernetes

Read-only review of how AI models are served on Kubernetes: startup timelines, reliability and access findings, in one HTML report.

Kubernetes-native control plane for multi-tenant agent platforms on EKS. A Platform CR is the tenant boundary: namespace, IAM identity, KMS grants, scoped datastore access. Budget, SLO and eval policies are CRs. Bedrock for models, Envoy AI Gateway for the data plane.

Inference-style API on Kubernetes: Helm chart, Prometheus error-budget alerts, Grafana dashboard, KEDA autoscaling on the requests each pod holds. Tested end to end on kind in GitHub Actions.

SRE incident agent served through a GPU control plane (guard, admit, place, queue, hop) on real vLLM, with measured findings.

Multi-cloud cost optimization and anomaly detection (FOCUS 1.0) for AWS, GCP & Azure — spend aggregation, forecasting, rightsizing, Prometheus/Grafana monitoring, and KEDA autoscaling.

Kubernetes operator and Python SDK for adaptive microservices. Compose graphs across clusters, enforce Cheeger bounds, and adapt topology, traffic and capacity to application demand.

Self-hosted LLM serving on EKS: vLLM, Karpenter GPU autoscaling, KEDA, and live cost per 1K tokens

jobq on Kubernetes: Terraform (kind + EKS), Helm, KEDA queue-lag autoscaling, CloudNativePG + PgBouncer, Gateway API, Prometheus SLO burn-rate alerts. A node killed mid-load loses 0 jobs.

Reusable Helm chart for multi-deployment Kubernetes applications — Deployments, Services, Ingress, CronJobs, Hooks, ExternalSecrets, RBAC, HPA, PDB, NetworkPolicy, and more.

Demo platform for elastic multi-cluster transaction processing across OpenShift, managed with RHACM and Argo CD

Provides a simple way to run Selenium Grid with Chrome, Firefox, and Edge using Container Platform, making it easier to perform browser automation at scale

SLOs for LLM inference on Kubernetes: TTFT, tokens/s, cost and quality SLIs with KServe, OpenTelemetry, Prometheus and KEDA.