Workflow showcase

Browse every example below by section: Simple examples, Machine Learning, Generative AI, Agentic workflows, and Scientific AI. Select a graph to open its page. Simple examples include runnable bundles or verified fixtures; the other sections describe workflow patterns and are not presented as executable bundles.

Simple examples

Simulation, real execution, and adapter fixtures. Each page states its prerequisites and validation scope.

prepare · edgeanalyze · cloudsummarize · edge

SimGrid

100 activities on 50 cores

Machine Learning

Training, tuning, inference, evaluation, and model delivery patterns. These are DAG designs, not verified executable bundles.

ingest datavalidatetrainevaluate

Training with MLflow

prepare dataset

trial 1trial 2trial N

compare trialsregister best

Hyperparameter search

fetch base modelprepare examplesfine-tune adaptersevaluate

publish adapterrecord lineage

LoRA fine-tuning

load modelpartition input

infer shard Ainfer shard Binfer shard N

merge predictionspublish

Batch inference

freeze evaluation set

evaluate Aevaluate Bevaluate C

compare metricsapprove candidate

Comparative model evaluation

load model

export ONNXexport TorchScript

quantize CPUquantize GPU

validate outputspublish variants

Model conversion and quantization

collect imagesimport labelsaugmenttrain detectorevaluatepackage model

Computer vision training

collect signalsbuild featurestrain detectorscore validationchoose thresholdpublish

Anomaly detection

prepare input

predict Apredict Bpredict C

combineevaluatepublish

Model ensemble

load teachersoft labelstrain student

qualitylatency

publish

Model distillation

freeze inputs

CPUGPUedge

normalizecompare cost and latency

Inference benchmark

profile workload

local GPUHPCcloud

compare plansselect scopetrain

Infrastructure selection for training

Generative AI

Data preparation and evaluation patterns for retrieval, generation, audio, and documents.

collect documentscleanchunkembedbuild index

retrieval testsanswer tests

publish

RAG index construction and evaluation

define schema

batch Abatch Bbatch N

filterdeduplicatequality reviewpublish

Synthetic dataset generation

ingest audiosegment

segment 1segment N

merge transcriptsummarizepublish

Audio transcription and summarization

ingest documents

OCRnative text

classifyextract fieldsvalidate schemapublish

OCR and document extraction

Agentic workflows

Bounded agent execution with explicit review, policy, and evidence stages.

prepare objectiveagent loopvalidate output

publish resultstore trace

Encapsulated agent task

build contextpropose actionpolicy checkhuman approvalexecute toolverify

Supervised tool loop

shared context

researchcodedomain

critic reviewsynthesizepublish evidence

Multi-agent review

Scientific AI

Simulation, analysis, surrogate models, and experiment feedback loops.

prepare experimentsimulationmeasurements

statisticsAI interpretation

validatereport

Scientific pipeline with AI interpretation

design samples

simulation Asimulation N

assemble datasettrain surrogatevalidatepublish

Surrogate model construction

observationsupdate modelpropose experimentssafety checkrun experimentsrecord observations

Model-guided experiment cycle

How to use a showcase

Choose a runnable example from Simple examples and open its repository checkout. The links below show the current examples on main; when using a packaged release, select its matching tag on Releases and use the examples from that tag.
Complete the scenario's prerequisites and API connection setup when using an existing server. The SLURM fixture starts its own server and sets its API URL in the procedure.
Submit the files in the stated order and wait for the run to finish.
Check the result listed on that scenario's page.

The files define the submitted model. If Desktop is connected to the same server, use it to inspect the saved records and result.

Example file index

Scenario	Files	What it demonstrates
Edge–cloud simulation	`examples/simulation`	Complete environment, topology, workflow, plan, and execution request
30 GB fan-out	`examples/simulation/30gb-fanout`	Shared links, parallel transfers, and fan-in
50-core fan-out	`examples/simulation/50core-fanout`	Per-core placement and parallel capacity
Kubernetes with Kind	`examples/kind`	Local real execution, namespace access, and requests
Hybrid local/Kind to GCP	`examples/real/hybrid-cloud-transfer`	Cross-runtime file handoff to an on-demand GCP worker; requires existing matching resources
Montage 58, GCP to local	`examples/real/montage-cloud-local`	Real Montage commands, twelve GCP `mProject` tasks, named FITS transfers, and local completion; observed run pending
Local direct	`examples/local/direct-hello`	A trusted host-local command and observed filesystem artifact
SLURM batch fixture	`examples/slurm/local-fixture`	Reproducible local validation of the SLURM adapter's batch submission and artifact path
Remote connections	`examples/connections`	SSH-based PlaFRIM, Santos Dumont, and Kind connections

Infrastructure configuration references

These files model infrastructure inputs, but are not end-to-end showcases. In particular, they do not include a portable cluster, a submitted run, or observed execution evidence.

Infrastructure	Files	Scope and next step
SLURM cluster	`examples/slurm`	A catalog for a partition, login node, shared storage, and runtime bindings. Pair it with an approved SLURM cluster and follow Connect an HPC and SLURM cluster before submitting a site-approved probe. The separate SLURM batch fixture validates only the local adapter path.