http://www.volundarhus.com/categories/mlops/Recent content in MLOps on VolundarhusHugoen-usMon, 20 Jul 2026 18:46:00 -0700http://www.volundarhus.com/posts/basic_mlops_pipeline/Mon, 20 Jul 2026 18:46:00 -0700http://www.volundarhus.com/posts/basic_mlops_pipeline/<p><a href="https://github.com/huntermills707/haterade" class="external-link" target="_blank" rel="noopener">GitHub Repo</a></p> <h2 id="why-i-built-this"> Why I built this <a class="heading-link" href="#why-i-built-this"> <i class="fa-solid fa-link" aria-hidden="true" title="Link to heading"></i> <span class="sr-only">Link to heading</span> </a> </h2> <p>Most MLOps tutorials stop at &ldquo;model in a notebook, Dockerfile around it.&rdquo; The hard part is everything after: how a trained artifact moves from a training run into a production-grade serving topology, how a new model version earns its way into production, and what it costs to serve. I wanted that whole path to be concrete and reproducible on hardware I own, so I built it: a text-toxicity classifier that goes train → register → export → serve → canary → promote (or roll back), with a feedback loop that routes user corrections back into training data.</p>