MLOps

Start here — the learning shelf#

Enough curated material to teach the whole discipline; the recurring pattern in your saves is “stop at the notebook is not enough.”

• awesome-mlops — the canonical curated list, and ml-ops.org, the same author’s site with ten MLOps books collected in one place.
• MLOps Zoomcamp (video) and its course page — DataTalks.Club’s free nine-weeks-to-production course, the most course-shaped thing here.
• 10 GitHub repositories to master MLOps · free courses across DS, DE, ML, MLOps and LLMOps · a free MLOps-basics ebook — the kdnuggets drip.
• Pau Labarta Bajo: one real-world project to learn MLOps — “stop reading, get your hands dirty”; the advice you keep re-saving in different forms.
• the 8-week ML, MLOps & Career Accelerator — a 2026 save, notable because it bundles the career transition itself into the curriculum.
• Oxford’s AI, generative AI, cloud and MLOps course — saved with a raised eyebrow at the C# in their agent examples.
• FastAPI for MLOps: project structure and API best practices — the serving layer, done properly.

MLOps is DevOps you already know#

The oldest thread in the tag, and the one that explains why you were early: this was DevOps people telling other DevOps people that ML was coming for them.

• Damian Brady: MLOps or DevOps for machine learning? and the sequel how does MLOps differ from DevOps? — the cleanest framing of the skills transfer you eventually made.
• Why should I care about MLOps? · How can MLOps improve my predictive models? — Microsoft’s One Dev Question shorts from 2020, when this still needed selling.
• MLflow + TensorFlow + Keras with PyCharm — your very first save under this tag, July 2018, barely a month after MLflow’s launch.
• Pulumi: the real AI challenge is cloud, not code — the thesis of your entire transition, in one title.
• Applying SRE principles to your MLOps pipelines — Google Cloud making explicit what you already suspected: error budgets and reliability thinking apply to model pipelines too.
• data drift, concept drift, model drift, observability, explainability — the monitoring vocabulary that separates ML in production from ML in a notebook.
• Philips’ MLOps platform on SageMaker — saved as “applying the DevOps mindset to Gen AI and ML”; a healthcare-scale proof it works.

The Azure shelf#

Thirteen links — the biggest cluster in the tag, tracking your Azure depth from AzureML infographics to full GenAIOps reference architecture.

• The early years: five best practices for the MLOps lifecycle with AzureML (2019) · MLOps: the path to building a competitive edge (2020) · a tale of two Azure Pipelines — integrating ML pipelines into an existing DevOps process.
• Hands-on: the Azure MLOps workshop from AzConf (video) and part 8 of the Azure MLOps Challenge blog.
• The charotamine pair — full IaC-flavored pipelines, your stack exactly: end-to-end MLOps with TensorFlow, Azure ML, GitHub Actions and Bicep · LLMOps with Azure AI, Prompt Flow, Bicep and GitHub Actions.
• Prompt Flow in anger: LLMOps with Azure ML Prompt Flow on an NER task · Operationalizing LLMs on Azure (Coursera).
• The reference material: GenAIOps for organizations with existing MLOps investments — the Azure Architecture Center page that answers the exact question an Enterprise AI Architect gets asked.
• Microsoft’s TechExcel LLMOps-automation workshop: the intro and Exercise 05: Automate Everything — plus the “code-first LLMOps” AI Tour demo repo, saved when the session was standing room only.

Google Cloud, Databricks & the platform stories#

• GenOps: the evolution of MLOps for gen AI and the GenAI MLOps blueprint for Vertex AI — Google’s answer to the Azure Architecture Center pages above; you collect these in matched pairs.
• how L’Oréal’s Tech Accelerator built its end-to-end MLOps platform — the enterprise case study.
• Databricks’ Big Book of MLOps, updated for generative AI — the free book that bridges both eras, and Feature & Function Serving: real-time structured data for RAG apps.
• Matt Turck’s MAD landscape 2024 — saved with your own periodization: “Gen 2 was MLOps”; AI developer platforms are Gen 3. Useful humility for a tag that peaked in 2024.

The LLMOps turn#

From late 2023 the tag pivots almost wholesale. Same discipline, new artifact: prompts and models-as-APIs instead of trained weights.

• LLMOps deployment architecture patterns — the August 2023 save where the turn begins.
• 5 essential steps to building LLM apps (video) · open-source LLMs, fine-tunes and RAG-based vector store APIs — the early stack-assembly phase.
• What Is LLMOps? — the O’Reilly report, saved on launch day.
• architecture & design principles for MLOps and LLMOps — the two disciplines treated as one, which is where you landed too.
• ZenML’s LLMOps database — hundreds of filterable industry case studies; you noted you’d been meaning to compile exactly this yourself.
• The Full MLOps/LLMOps Blueprint — the consolidated 2025 overview.
• the free hands-on LLM course — real-world LLM apps with MLOps best practices baked in.
• an end-to-end MLOps pipeline with open-source tools — worth keeping around as the vendor-neutral counterweight to all the cloud-platform links above.

Where it’s heading: evals and agent ops#

The newest saves suggest the discipline’s next rename — and they point straight at your agentic-AI work.

• eval-driven development in MLflow — find failures with evals, fix, rerun the same suite, ship with tracked deltas. CI/CD logic applied to model behavior; the most architecturally important link on this page.
• agent evaluations in your LLMOps with GitHub Actions — evals wired into the pipeline you already know how to run.
• “the holy grail of MLOps was always continual learning” — your saved note observes the same concepts resurfacing in agent building as “agent ops.” That one sentence is the bridge between this tag and AI Agents.
• GraphRAG analysis part 2: graph vs. vector retrieval — the MLOps community turning its rigor on RAG architecture choices.
• Chrome’s built-in AI APIs — Gemini Nano in the browser — the contrarian save: sometimes the best model deployment is none at all.

All links on this topic#