Building Artificial Intelligence & Machine Learning Applications on Kubernetes

Dive into end-to-end AI/ML on Kubernetes: recap core k8s primitives, explore the Kubeflow control and data plane, and master installation via kfctl or Operator. Launch and manage Jupyter notebooks with PVC and S3/MinIO attachments, customize environments with GPU quotas, and build your own ML-ready container images. Orchestrate training, tuning and serving with Pipelines, Katib and KServe, all secured behind Dex/OIDC multi-user isolation. Hands-on labs guide you through CRDs, GitHub-backed authentication and real-world MLOps workflows.

Find Our Students At

About

Topics

Testimonials

About the course

Accelerate your AI/ML projects by learning how Kubernetes empowers reproducible, scalable model development and deployment. Start with a foundations recap—Pods, Services, ConfigMaps—and progress to Kubeflow components: Notebooks, Pipelines, Metadata, Katib and KServe. Install and verify Kubeflow, configure GitHub-based Dex authentication, and enforce per-user resource quotas. Through labs you’ll spin up JupyterLab servers, attach persistent storage, build custom images, and deploy production-grade inference services.

Course Content

Foundations & Prerequisites

Kubernetes recap: Pods, Services, Deployments, ConfigMaps, Secrets
Machine Learning Overview
Artificial Intelligence Overview
Role of Kubernetes in AI & ML

Kubeflow Architecture & Installation

Kubeflow components overview (Notebooks, Pipelines, Metadata, Katib, KServe)
Lab: Explore Kubeflow CRDs in your cluster
Lab: Install Kubeflow via kfctl (or Operator) and verify core pods
Authentication (Dex/OIDC) & multi-user isolation
Lab: Configure Dex with a GitHub connector

Jupyter Notebooks & Workspace Management

Kubeflow “Notebook” CRD & lifecycle
Lab: Launch a Notebook server from the UI
Underlying k8s resources (PVCs, ServiceAccounts)
Lab: Attach a PVC; mount a MinIO/S3 bucket
Customizing user environments (images, GPU quotas)
Lab: Build and register a custom JupyterLab image with pre-installed ML libraries

Pipelines & Workflow Orchestration

KFP SDK & DSL concepts
Lab: Write a 3-step pipeline (preprocess → train → eval) in Python DSL
ContainerOp vs. pre-built components
Lab: Compile & upload the pipeline; run it and inspect logs
Pipeline parameters, artifacts, and caching
Lab: Enable experiment tracking and pipeline caching

Hyperparameter Tuning with Katib

Katib CRD model: Experiment, Trial, Suggestion
Lab: Define a KatibExperiment for your training container
Search algorithms (Grid, Random, Bayesian)
Lab: Run a Bayesian optimization; view trial results in the UI
Metrics collection & early-stopping
Lab: Extract the best hyperparameters and retrain your model

Model Serving (KServe)

InferenceService CRD: predictor, explainer, transformer
Lab: Package a trained model (SavedModel or joblib) into S3/MinIO
Autoscaling & GPU vs. CPU resource configs
Lab: Deploy an InferenceService and test with curl or Python
Canary & blue-green rollouts
Lab: Configure traffic splitting for A/B testing

Metadata, Tracking & Pipelines UI

ML Metadata (MLMD) fundamentals
Lab: Visualize pipeline runs and artifact lineage in the UI
Artifact lineage and reproducibility
Lab: Hook an MLflow server into your pipeline for metrics logging
Integrating MLflow or Weights & Biases
Lab: Query past runs via the Metadata API

Security, RBAC & Multi-Tenant Isolation

k8s RBAC essentials
Lab: Create separate namespaces for teams with granular RBAC
Namespaces, NetworkPolicies, and ResourceQuotas
Lab: Apply NetworkPolicies to isolate notebook traffic
Securing MinIO/S3 and Ingress TLS
Lab: Issue certs via cert-manager and secure the Ingress

Monitoring, Logging & Alerting

Prometheus metrics (Pod, GPU, pipeline durations)
Lab: Scrape Kubeflow and custom app metrics in Prometheus
Lab: Build a Grafana dashboard showing pipeline health
Centralized logs (EFK or Loki) and alert rules
Lab: Ship logs from notebooks and pipelines to Elasticsearch or Loki

What our students say about us

Frequently Asked Questions

Is this training live or pre-recorded?

The training is live instructor led training which is available in classroom as well as online format. We also record every training session which is then uploaded to our student portal.

How will I join the live online training?

The live online training is conducted via the zoom software, we will be providing you with the zoom meeting link to join the training.

How many students are there in a single batch?

On an average one batch will have a maximum of 18 students. We keep smaller batch sizes to promote interaction between the students and the instructor.