AI-Driven Kubernetes for GitOps and Agent Systems
Mode: Online
Workshop Duration: Duration: 2 Days ~ 4 hours each session
Dates: 11th & 12th Sep 2025
Timings: 2:00 PM – 6:00 PM IST (each day)
Overview:
This intensive workshop is meticulously designed for experienced technology professionals with prior Kubernetes knowledge who want to master advanced Kubernetes use cases.
You will gain practical, hands-on experience deploying, managing, and scaling AI/ML workloads in a GitOps framework, and explore how Kubernetes can serve as a runtime for agentic AI systems.
Key Objectives:
By the end of the workshop, participants will:
✅ Deploy and scale GenAI and ML workloads on Kubernetes
✅ Implement observability and autoscaling with custom Prometheus metrics
✅ Package and deliver workloads using Helm and ArgoCD
✅ Extend Kubernetes with Operators and Custom Resource Definitions (CRDs)
✅ Run Kubernetes as a runtime for agentic AI systems
✅ Apply Kubernetes-native support for GPUs and AI enhancements
Prerequisites:
- Strong understanding of Kubernetes fundamentals (Pods, Services, Deployments, YAML)
- Hands-on experience with containerized applications and microservices
- Familiarity with Git and CLI tools
- Basic knowledge of AI/ML models and containerized APIs
Course Outline
Day 1 (4 Hours)
Module 1: Deploying GenAI APIs on Kubernetes
Run a containerized LLM inference service on Kubernetes
- FastAPI-based GenAI app (e.g., summarizer/chatbot)
- Write YAML specs (Deployments, Services, Ingress)
- Create base Helm chart for the app
Hands-on Lab**:
- Deploy FastAPI+Ollama model on KIND with Ingress
- Test endpoint with curl or Postman
Module 2: Observability + Autoscaling for AI APIs
Monitor and auto-scale AI APIs
- Prometheus + Grafana stack via Helm
- Export FastAPI metrics using Prometheus Middleware
- Configure Prometheus Adapter for custom metrics
- Use HPA with request count or latency
Hands-on Lab**:
- Visualize LLM metrics
- Set up autoscaler to respond to load
Module 3: GitOps Deployment with Helm + ArgoCD
Automate GenAI delivery across environments
- Helm values for staging vs production
- Use Kustomize for overlays
- ArgoCD (or FluxCD) for GitOps-based syncing
- Argo Image Updater for CI/CD triggers
Hands-on Lab**:
- Deploy GenAI app to two environments via GitOps
- Trigger new release by updating Docker tag
Day 2 (4 Hours)
Module 4: AI-Augmented Kubernetes with kubectl-ai + MCP
Use LLM-powered agents to *assist* platform users with operational tasks, YAML generation, error resolution, and lifecycle automation.
- What is AI-Augmented Kubernetes?
- Introduction to `kubectl-ai`
- Kubernetes MCP Servers
Hands-on Lab**:
Using `kubectl-ai`
Deploying Kubernetes MCP Server
Module 5: Deploying Agentic AI Systems on Kubernetes
Run agent-based workflows (not tied to Argo)
- Architecture of agentic systems on K8s
- LangGraph-based multi-agent app:
- Nodes = Agents (Docker containers or Jobs)
- Edges = Communication via queues or volume
- Optional: Use Redis, RabbitMQ, or Kubernetes Jobs as workflow steps
- Use Job, CronJob, StatefulSet, and sidecar patterns for agent deployment
Hands-on Lab**:
- Deploy a LangGraph agent workflow (build/test/deploy pipeline)
- Agents: Code Generator, Reviewer, Image Builder Scanner (Trivy) Publisher
Module 6: Native Kubernetes AI/ML Features (Showcase + Demos)
Demo + Visuals**:
- How GPU requests work via device plugin
- How to schedule a training job using JobSet + Kueue