Polyaxon v3 is coming →

Learn how to run AI workloads in production.

Follow practical paths through GPU infrastructure, MLOps, sandboxes, evaluation, observability, and workload automation.

GPU orchestration

What is GPU orchestration?

Understand how orchestration connects workflows, queues, and resource placement.

MLOps & tracking

ML Experiment Tracking: What to Log and How to Compare Runs

Record and compare parameters, metrics, and outputs.

Sandboxes

Start a sandbox

Create a sandbox-enabled run and connect with Python or the CLI.

LLM evaluation

How to evaluate RAG systems

Measure retrieval and generation separately, then assess the complete task.

Schedule and operate shared GPU workloads.

Queues, placement, recovery, and utilization for shared GPU clusters.

Explore GPU orchestration

Choosing GPU infrastructure for Polyaxon

Shortlist Lambda, CoreWeave, Nebius, Crusoe, and Vast.ai by Kubernetes fit, capacity, infrastructure ownership, and workload requirements.

What is GPU orchestration?

Understand how orchestration connects workflows, queues, and resource placement.

Queue management for machine learning workloads

Set priorities and concurrency for shared infrastructure.

Gang scheduling for distributed training

Coordinate worker admission for distributed training.

How to improve GPU utilization

Find queue delays, input bottlenecks, and fragmented capacity.

Make experiments and models reproducible.

Tracking, metadata, registries, and repeatable model development.

Explore MLOps & tracking

What is MLOps?

Understand operations across the model development lifecycle.

MLOps core principles

Connect collaboration, automation, version control, and monitoring.

ML Experiment Tracking: What to Log and How to Compare Runs

Record and compare parameters, metrics, and outputs.

ML Metadata Store: Examples, Artifacts, and Lineage

Preserve the context and lineage behind your results.

Develop in the same environment as your workloads.

Interactive environments, access, resource control, and automation.

Explore Sandboxes

What is an AI sandbox? Development environments and execution isolation

Distinguish interactive environments from runtime isolation and define their access.

Start a sandbox

Create a sandbox-enabled run and connect with Python or the CLI.

Connect with SSH

Connect local terminals and development tools to your run.

Automate with Python and CLI

Automate process and file operations from Python or the CLI.

Evaluate AI changes before release.

Metrics, judges, regression suites, and production feedback.

Explore LLM evaluation

How to evaluate RAG systems

Measure retrieval and generation separately, then assess the complete task.

How to evaluate AI agents

Evaluate task outcomes, tool use, trajectories, latency, and cost.

LLM-as-a-judge: Design and validate model-based evaluators

Design rubrics and validate model-based scores against human decisions.

Offline vs. online evaluation for generative AI

Connect reproducible release checks to production feedback.

Trace and evaluate agent behavior.

Model calls, tools, handoffs, outcomes, and evaluation across the agent lifecycle.

Explore Agents

The AI agent lifecycle

Explore the development, evaluation, and observation loop.

How to evaluate AI agents

Define success for outcomes, actions, and efficiency.

AI agent tracing: How to debug tools, loops, and handoffs

Follow model calls, tool use, loops, and handoffs.

MCP observability: Monitor tools, resources, and context

Inspect context and operations across MCP clients and servers.

Operate AI workloads on Kubernetes.

Architecture, storage, metrics, resource inspection, and failure diagnosis.

Explore Kubernetes for AI

Kubernetes architecture for ML workloads

Understand control planes, workers, services, and storage.

Kubernetes storage classes overview

Learn dynamic provisioning, volume binding, and persistent claims.

How to leverage Kubernetes metrics

Choose signals for resource usage, cluster state, and debugging.

How to use kubectl describe

Investigate resource state, configuration, and events.

Connect runtime signals to AI outcomes.

Traces, metrics, evaluations, and deployment context for diagnosing failures.

Explore AI observability

What is AI observability?

Connect traces, metrics, evaluations, and runtime context.

AI agent tracing: How to debug tools, loops, and handoffs

Debug tool calls, loops, and handoffs.

MCP observability: Monitor tools, resources, and context

Follow tools, resources, and context across clients and servers.

Turn production traces into regression tests

Turn representative failures into reproducible evaluation cases.

Operate LLM applications in production.

Prompt versions, model access, cost, evaluation, and release decisions.

Explore LLMOps

What is LLMOps? From prototype to production

Understand development, evaluation, deployment, and feedback.

Prompt versioning for production AI systems

Treat prompts as versioned artifacts with evaluation and rollback.

What is an AI gateway?

Understand routing, resilience, policy, and model access.

LLM cost monitoring: Measure cost per successful task

Relate model and tool costs to successful task outcomes.

Test AI systems before failures reach users.

Authorized adversarial cases, evidence capture, and regression testing.

Explore AI red teaming

What is AI red teaming?

Define test boundaries, adversarial cases, and evidence for remediation.

How to test prompt injection in LLM applications

Build direct and indirect injection cases and inspect actual tool actions.

How to red team AI agents

Test permissions, memory, handoffs, retries, and execution budgets.

Continuous AI red teaming in CI/CD

Preserve reviewed findings as complete, repeatable release checks.

Automate repeatable AI workflows.

Dependencies, retries, schedules, hooks, and repeatable execution.

Explore Pipelines & automation

Pipelines and orchestration

Understand pipelines, orchestration, and recurring workflows.

Automation with DAGs

Express dependencies and control execution with a DAG.

Running on schedule

Run components and workflows on a recurring schedule.

Executing Hooks

Trigger follow-up operations when a run finishes.

Link model versions to their evidence.

Models, runs, artifacts, lineage, and approval context in one path.

Explore Model registry & lineage

Registries and reusable assets

Understand how Polyaxon organizes reusable ML assets.

Model Registry

Organize model assets and their versions.

Model versions

Associate model versions with the runs that produced them.

Run Lineage

Inspect the input and output dependencies of a run.