# Choosing your next step by role

You can now open a session, launch a [cluster](/glossary/cluster), dispatch parallel work, and collect results. That is the
core loop.

Where to go next depends on what you are actually trying to accomplish.

## If you analyze data

You want [Distributed Python with `eugo.hpc`](/courses/distributed-python-with-eugo-hpc). It covers
[fan-out](/glossary/fan-out) properly, per-call resource requests, actors for stateful work, and a realistic parallel
pandas workload.

Then [scaling a real workload](/courses/scaling-a-real-workload) for the case where the dataset is
large enough that I/O becomes the constraint.

The [data scientist path](/learning-paths/data-scientist) sequences these for you.

## If you train or serve models

Go to [GPU acceleration on Eugo](/courses/gpu-acceleration-on-eugo). It covers which operations reach
the GPU, how to request one, and how to confirm it happened.

Pair it with [how automatic optimizations work](/courses/how-automatic-optimizations-work). Knowing
what the runtime does on your behalf is what lets you write code it can help.

The [ML engineer path](/learning-paths/ml-engineer) covers both.

Remember: **PyTorch, not TensorFlow.** TensorFlow, spaCy, and JAX are not supported.

## If you are setting up for a team

[Workspaces & organizations](/courses/workspaces-and-organizations) is the one you need. It covers
[organizations](/glossary/organization), [workspaces](/glossary/workspace), members, roles, plans, and the activity trail.

The [workspace admin path](/learning-paths/workspace-admin) adds the platform context around it.

## If you want the full tour first

[Eugo 101](/courses/eugo-101) covers the whole platform in order rather than by role. Worth it if you
would rather have the complete map before specializing.

## If you just want to see the numbers

The [interactive demos](/interactive) let you explore scaling curves, a CPU-versus-GPU comparison
including the cases where a machine's GPU loses to its own CPU, and a cluster sizer that estimates
worker count from the shape of your workload. The chart figures there are illustrative rather than
measured, and the page says so — they are for reasoning about a regime, not for quoting.

## Two things worth reading regardless

- [Workload readiness checklist](/resources/workload-readiness-checklist), to verify a workload is
  worth distributing before spending cluster time on it.
- [Glossary](/glossary), the vocabulary used consistently across these courses and the platform
  docs.

---

Source: https://university.eugo.io/lesson/getting-started-with-eugo/next-steps