The latest articles on Forem by Murat (@flyingriverhorse). https://forem.com/flyingriverhorse https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F3674852%2Fce78c019-b061-4fb7-8244-842c2a343a4e.jpg https://forem.com/flyingriverhorse en Murat Tue, 23 Dec 2025 09:53:56 +0000 https://forem.com/flyingriverhorse/from-feature-engineering-to-deployment-a-local-first-mlops-workflow-with-skyulf-140g https://forem.com/flyingriverhorse/from-feature-engineering-to-deployment-a-local-first-mlops-workflow-with-skyulf-140g <p>Most ML tooling assumes your data can live in someone else’s cloud, or that your team wants to assemble a stack of separate tools (orchestrator + tracking + deployment + UI) and spend weeks wiring everything together.</p> <p><strong>Who Skyulf is for</strong><br> Skyulf is built for:</p> <ul> <li>Teams working with sensitive/regulated data</li> <li>People who want a <strong>local-first</strong> workflow (laptop → server → on-prem)</li> <li>ML engineers and data scientists who prefer <strong>one integrated workflow</strong> over a pile of disconnected components</li> <li>Anyone iterating quickly on models and wanting workflows that stay <strong>visible, repeatable, and easy to review</strong>.</li> </ul> <p><strong>What you can do with Skyulf</strong><br> Skyulf focuses on the end-to-end loop:</p> <ol> <li> Ingest + explore data</li> <li> Feature engineering (visually, as a pipeline)</li> <li> Training (including background jobs)</li> <li> Deployment (self-hosted inference service)</li> <li> Verification with an API testing panel (send JSON, view response/latency)</li> </ol> <p>pipeline → run → deploy → test API</p> <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fs6fcjtkg8ag3afhk3ftq.jpg" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fs6fcjtkg8ag3afhk3ftq.jpg" alt=" " width="800" height="340"></a></p> <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2F8yh9r25t0s8028ubek15.jpg" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2F8yh9r25t0s8028ubek15.jpg" alt=" " width="800" height="345"></a></p> <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2F10t6gio1s4cgvm42ek2c.jpg" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2F10t6gio1s4cgvm42ek2c.jpg" alt=" " width="800" height="337"></a></p> <p><strong>Why “visual pipelines” matter (beyond aesthetics)</strong><br> A visual pipeline canvas isn’t just a pretty UI; it’s a way to make ML workflows:</p> <ul> <li>explainable (anyone can see what happens between raw data and model)</li> <li>repeatable (less tribal knowledge, fewer hidden scripts)</li> <li>reviewable (pipelines become artifacts you can share and iterate on)</li> </ul> <p><strong>What’s next</strong><br> Skyulf is open source and evolving. Near-term focus areas:</p> <ul> <li>more example pipelines (tabular, time-series, text/embeddings)</li> <li>more models</li> <li>better packaging for “one command” self-hosting</li> <li>integrations/export paths for teams already using other tools.</li> </ul> <p><u>If you want to try it, start here:</u></p> <p>GitHub repo: <a href="https://github.com/flyingriverhorse/Skyulf" rel="noopener noreferrer">https://github.com/flyingriverhorse/Skyulf</a><br> Website/docs: <a href="https://www.skyulf.com/" rel="noopener noreferrer">https://www.skyulf.com/</a></p> <p>If you only want the Python engine (no UI), for example, to integrate Skyulf into your own application or scripts, you can install skyulf-core directly via pip:</p> <p><code>pip install skyulf-core</code></p> <p>If you run it and have feedback, open an issue, especially around onboarding and docs clarity.</p> mlops opensource datascience machinelearning