Skip to main content

Showcases

Explore 28 executable workflows organized by technical domain. Each page explains the workflow DAG, activity responsibilities, inputs, outputs, provenance, and execution considerations without embedding implementation code in the documentation.

The complete Dockerfiles, Python programs, generators, input fixtures, runners, and validators are maintained separately in UFFeScience/akoflow-examples. A single source-bundle link on each page connects the explanation to its versioned implementation.

Choose a category​

  • Scientific Workflows — the full Montage 050d astronomical mosaic, numerical analysis, surrogate models, and experiment feedback loops.

  • Machine Learning — training, tuning, inference, evaluation, conversion, vision, ensembles, and infrastructure selection.

  • Generative AI — retrieval, synthetic data, audio transcription, document extraction, and offline video creation.

  • Document AI — classification, structured extraction, and reconciliation.

  • Media intelligence — reproducible video inspection, segmentation, audio analysis, and timelines.

  • Rendering — procedural scenes, parallel frame shards, and animation composition.

  • Agentic workflows — bounded agents, supervised tool use, and multi-agent review.

Scientific Workflows

Executable simulation, analysis, surrogate-model, and experiment-feedback workflows.

Montage 050 locally

Scientific pipeline with AI interpretation

Surrogate model construction

Model-guided experiment cycle

Machine Learning

Executable training, tuning, inference, evaluation, and model-delivery workflows.

Training with MLflow

Hyperparameter search

LoRA fine-tuning

Batch inference

Comparative model evaluation

Model conversion and quantization

Computer vision training

Anomaly detection

Model ensemble

Model distillation

Inference benchmark

Infrastructure selection for training

Generative AI

Executable data preparation and evaluation workflows for retrieval, generation, audio, and documents.

RAG index construction and evaluation

Synthetic dataset generation

Audio transcription and summarization

OCR and document extraction

AI Video Studio

Local RAG Factory

Document AI

Executable extraction and reconciliation workflows over auditable document corpora.

Document Intelligence

Media intelligence

Offline audiovisual inspection, segmentation, measurement, and timeline workflows.

Video Intelligence

Rendering

Procedural scene construction and parallel frame rendering workflows.

Blender Render Farm

Agentic workflows

Executable bounded-agent workflows with explicit review, policy, and evidence stages.

Encapsulated agent task

Supervised tool loop

Multi-agent review

Validation model​

Each page includes a copyable local command, its prerequisites, a precise description of the portable workload, and its expected output files. The 27 small examples execute real CPU computations and have both standalone local validation and completed Docker-backed daemon runs. The evidence includes all 171 activity captures and the 27 observed DAGs, with links to their measured results and execution records.

The local profiles are deliberately small and explicit: for example, geometric images instead of CIFAR-10, a rank-one adapter on a tiny matrix instead of an LLM, TF-IDF retrieval instead of hosted embeddings, and a synthetic tone vocabulary instead of natural-speech recognition. GPU, edge, and infrastructure comparison profiles distinguish CPU measurements from declared placement assumptions. The training example uses the actual MLflow SDK and a local SQLite tracking store.

Montage 050 is the full 58-activity scientific workflow and uses the pinned original image. Its image and command contract are verified separately from its long-running execution. A started run is shown as running until its final files are validated.

Read the verification records for the complete catalog. Provider-specific execution on HPC, Kubernetes, or cloud infrastructure still requires the environment described by the corresponding infrastructure guide.