I’ve endeavored to create a visualization of the MLOps framework that spans from Data Engineering to Model Monitoring and stitching over various components and workflows.

This diagram is crafted not just for the tech-savvy, but it is also catered towards a general audience who wish to gain a deeper understanding of MLOps. (I know it is not well crafted as diagram moving balls, but next time I will try to fix this 🙂 )

In an attempt to make this even more accessible, I’ll strive to explain the components in a simplified, user-friendly manner so that everyone can grasp the concepts easily.

Version Control: Imagine you’re painting a picture. Each time you make changes to your painting, you take a picture of it. That’s what version control does — it keeps a picture of each change you make to your code. When you’re done with the changes, you say, “I’m done!” That’s like merging your code to the main branch, and it triggers the next steps.

Feature Preprocessing: Now, imagine you’re cooking. Before you start cooking, you have to prepare your ingredients — wash them, chop them, etc. That’s what feature preprocessing does — it prepares your data (the ingredients) to be used by the machine learning model (the recipe).

Model Training and Validation: This is like the actual cooking part. You use your recipe (the model) to cook the ingredients (the data). Once the dish (the model’s output) is ready, you taste it to make sure it’s good. This is what we call validating the model.

Model Registry and Packaging: When the dish is tasty, you pack it nicely in a lunchbox (that’s the model registry) to take it to school (deployment). And you wrap it (containerize it) so it doesn’t spill or get ruined.

Experiment Tracking and Deployment: Think of it as writing down what ingredients and steps you used to cook the dish. So if someone else (or you, later) wants to cook it again, they know exactly what to do. And once your lunchbox is packed, you get on the school bus (deployment pipeline) to take your lunch (model) to school (production).

Inference Request: This is like your friends at school asking to taste your dish. They send a request, and they get to taste your dish.

Feature Store: Think of it as a fridge where you store all your ingredients. Some ingredients are ready to use (static features), and some you have to prepare each time (dynamic features).

Model Monitoring: This is like checking your dish from time to time to make sure it’s not gone bad. If it has (model quality degrades), you cook it again (retrain the model).