Agentic AI Private

Deploying Agents on Kubernetes, Databricks, and Snowflake (AGNTDV105)

3 days
Agentic AIKubernetesDatabricksSnowflakeAI

Deploy production AI agents on Kubernetes, Databricks, and Snowflake with autoscaling, governance, CI/CD, cost control, and security review readiness.

Talk to a Learning ConsultantRequest a Private Class

Register or Request Training

  • Private class for your team
  • Live expert instructor
  • Online or on‑location
  • Customizable agenda
  • Proposal responses same day as request
Request Proposal

Course Overview

An agent on a laptop and an agent serving an organization are different systems. Production agents are long-running, stateful, bursty, expensive, and sitting directly on top of governed enterprise data — which is why deployment decisions usually come down to where the data lives.

This course covers three realistic targets and the reasoning that selects between them. On Kubernetes, students containerize and deploy an agent service with autoscaling, persistent checkpoint state, secret management, GPU-backed self-hosted inference, and streaming through ingress. On Databricks, they deploy agents inside the lakehouse with Unity Catalog governance, vector search, model serving, and MLflow-tracked agent lifecycles. On Snowflake, they build agents that run next to the warehouse with Cortex, Snowpark Container Services, and role-based access control.

Throughout, the course keeps returning to the operational realities: cost control, latency, scaling on non-uniform workloads, data governance, network egress, failure recovery, and what it takes to pass a security review.

Who should attend: Platform and DevOps engineers, ML engineers, and data platform engineers responsible for running agents in production.

Course Benefits

  • Choose a deployment target based on data gravity, governance, latency, and cost
  • Package an agent as a production service with health checks, graceful shutdown, and streaming
  • Deploy on Kubernetes with autoscaling, persistent state, secrets, and observability
  • Self-host inference on GPU nodes and decide when that beats a hosted API
  • Deploy and govern agents on Databricks with Unity Catalog, Vector Search, Model Serving, and MLflow
  • Build and deploy agents on Snowflake with Cortex, Snowpark Container Services, and RBAC
  • Implement CI/CD, environment promotion, and safe rollback for agent systems
  • Control and attribute cost across inference, retrieval, and compute
  • Meet security and compliance requirements: network isolation, data residency, audit, and secret handling

Delivery Methods

Private Class
Delivered for your team at your site or online.

Course Outline

  • Day 1 — Kubernetes
    • Module 1: Choosing a Deployment Target
      • Agent workload characteristics: long requests, streaming, bursty concurrency, stateful runs
      • Data gravity as the deciding factor; governance and residency constraints
      • Hosted model APIs vs. self-hosted inference: cost, latency, control, and compliance
      • Reference architectures for the three targets, with honest trade-offs
    • Module 2: Agents on Kubernetes
      • Deployments, services, and ingress for long-lived streaming requests
      • Configuration and secrets; external secret stores
      • Resource requests, limits, and the memory profile of agent processes
      • Checkpoint state: managed Postgres, StatefulSets, and session affinity
      • Queues and workers for long-running or asynchronous agent runs
    • Module 3: Scaling, Reliability, and Cost on Kubernetes
      • HPA and KEDA: scaling on concurrency and queue depth rather than CPU
      • Cold starts, warm pools, and request timeouts through the whole path
      • Rate limits, backpressure, and retry budgets against model providers
      • Multi-tenancy, namespace isolation, and network policy with egress control
      • Self-hosted inference: GPU node pools, vLLM, batching, and utilization economics
  • Day 2 — Databricks
    • Module 4: Agents in the Lakehouse
      • Databricks architecture for agent workloads; where compute and data meet
      • Unity Catalog as the governance boundary: data, functions, and models
      • Unity Catalog functions as governed agent tools
      • Vector Search: index types, sync modes, and permission inheritance
    • Module 5: Agent Lifecycle and Serving on Databricks
      • MLflow for agent logging, tracing, versioning, and evaluation
      • Model Serving endpoints: scale-to-zero, provisioned throughput, and routing
      • External models and gateway patterns for provider access
      • Jobs, workflows, and scheduled or batch agent workloads
      • Secrets, service principals, and on-behalf-of-user authorization
  • Day 3 — Snowflake, CI/CD, and Operations
    • Module 6: Agents on Snowflake
      • Snowflake architecture for AI workloads; warehouses, roles, and credit economics
      • Cortex functions and LLM access inside SQL
      • Cortex Search and semantic models for retrieval over governed data
      • Role-based access control as the agent's permission model; secure views and row access policies
      • Snowpark Container Services for custom agent runtimes
    • Module 7: CI/CD and Promotion
      • Environment strategy: dev, staging, production across all three targets
      • Pipelines for code, prompts, indexes, and models as separate release artifacts
      • Evaluation gates in the deployment pipeline
      • Blue/green and canary rollout; rollback that actually works for stateful agents
      • Infrastructure as code for agent platforms
    • Module 8: Running Agents Day to Day
      • Unified observability across platforms; exporting traces from managed services
      • SLOs for agent systems and what to page on
      • Cost attribution per tenant, per feature, and per step; budget enforcement
      • Incident response: debugging a bad agent run from trace to root cause
      • Security review checklist: isolation, egress, secrets, audit, data residency, model governance

Class Materials

Each student receives a comprehensive set of materials, including course notes and all class examples.

Class Prerequisites

Experience in the following is required for this Agentic AI class:

  • Container and Kubernetes fundamentals
  • Python proficiency
  • Cloud experience
  • A working agent application, such as from AGNTDV101 through AGNTDV104 in this series, or equivalent

Experience in the following would be useful for this Agentic AI class:

Prior Databricks or Snowflake exposure is helpful for Days 2–3 but not assumed.

Have questions about this course?

We can help with curriculum details, delivery options, pricing, or anything else. Reach out and we’ll point you in the right direction.

Talk to a Learning Consultant