Lecture ( by Vikash Das, youtube ) →

https://youtu.be/5pniK1RV_6o?si=ABDUyVDs620AjWAM

NOTES:

image.png

Basics you should have before starting MLOps → what each point means:

  1. Python basics + VSCode + Anaconda → Be comfortable writing simple Python. VSCode is a code editor — the app where you actually write and run your code. Anaconda manages Python environments: think of an environment as a separate clean box holding one Python version + the exact libraries a project needs, so two projects don't clash. Short version →
  2. Data Science project lifecycle (hands-on) → You've built at least one project end-to-end: get data → clean it → train a model → check results. MLOps is basically about automating and managing this same flow, so you need to have done it manually first.
  3. Machine Learning understanding → A basic grip on how ML models work (training, testing, accuracy, overfitting). MLOps doesn't teach ML — it runs and maintains ML models in the real world, so the ML foundation has to already be there.
  4. GitHub account + a cloud account (AWS / Azure / GCP) → GitHub stores your code online and tracks versions/history. AWS (Amazon), Azure (Microsoft), GCP (Google) are the three big cloud providers — rented computers/servers on the internet where you run and deploy your project instead of your own laptop. They're competitors doing the same job, so you only need one of the three.

image.png

Introduction to MLOps:

Top half (the definition):

MLOps = "Machine Learning Operations." It's a set of practices that automate and simplify the whole journey of an ML model → from building it to running it live. (Deployment = putting your model out in the real world where users actually use it.) The name mixes ML + DevOps, where DevOps is the existing set of practices software teams use to ship and run normal apps smoothly. So MLOps borrows that DevOps discipline and applies it to ML models, so they get built and deployed in a reliable, repeatable way instead of being stuck in a Jupyter notebook.

In one line → MLOps brings order and automation to building, shipping, and maintaining ML models.

Bottom half (the diagram):

The infinity-loop shows the MLOps lifecycle as 3 connected stages that flow into each other. It's drawn as a loop on purpose → you keep coming back and improving, it's not a one-time straight line.