From feature engineering to deployment: a local-first MLOps workflow with Skyulf

Source: Dev.to

Who Skyulf is for

• Teams working with sensitive/regulated data
• People who want a local‑first workflow (laptop → server → on‑prem)
• ML engineers and data scientists who prefer one integrated workflow over a pile of disconnected components
• Anyone iterating quickly on models and wanting workflows that stay visible, repeatable, and easy to review

What you can do with Skyulf

• Ingest + explore data
• Feature engineering (visually, as a pipeline)
• Training (including background jobs)
• Deployment (self‑hosted inference service)
• Verification with an API testing panel (send JSON, view response/latency)

pipeline → run → deploy → test API

Why “visual pipelines” matter (beyond aesthetics)

• Explainable – anyone can see what happens between raw data and model
• Repeatable – less tribal knowledge, fewer hidden scripts
• Reviewable – pipelines become artifacts you can share and iterate on

What’s next

• More example pipelines (tabular, time‑series, text/embeddings)
• More models
• Better packaging for “one command” self‑hosting
• Integrations / export paths for teams already using other tools

Getting started

• GitHub repo:
• Website:

Install the Python engine only

If you only want the Python engine (no UI) to integrate Skyulf into your own application or scripts:

pip install skyulf-core

Contribute

If you run it and have feedback, open an issue, especially around onboarding and docs clarity.