Free resources that I followed to master MLOps in 2025
My list of trusted resources that I used to master MLOps. Sharing them here with you all, along with links of every Resources.
Hello Everyone
Welcome to your AKVAverse, I’m Abhishek Veeramalla, aka the AKVAman, your guide for Cloud, DevOps, and AI.
When I started learning MLOps, I felt lost.
There was no single place to learn everything step by step. Some blogs only explained theory, some courses taught just one tool, and many tutorials expected you to already know a lot.
So I had to learn the hard way, trying different resources, leaving the ones that didn’t help, and sticking to the ones that actually worked for me.
As promised in our YouTube video, I have put everything together in this newsletter. Here, I’m sharing the same steps and resources I used to learn MLOps in the right order, with all the links in one place.
1. DevOps Fundamentals
You can’t get good at MLOps without first learning DevOps. Why? Because MLOps is basically DevOps for machine learning.
Here is what I practiced:
Linux: Used daily until it felt natural.
Git: Practiced branching and merging every day.
Docker: Broke many containers before understanding images and volumes.
Kubernetes: Tricky at first, but deploying your app made it clear.
CI/CD: Began with Jenkins, later explored GitHub Actions.
Terraform: Treated infrastructure as code from the start.
AWS: Started with manual setup, then automated using Terraform.
Python: Served as the glue language connecting everything.
Link of the Resources:
DevOps YouTube Playlist
2. ML Fundamentals
At first, I thought I could skip ML theory and directly jump to MLOps. Big mistake.
When you don’t understand the ML lifecycle, you don’t know what you are automating.
So I forced myself to learn the basics:
Model development: Learned how to build and train simple models.
Experiment management: Practiced tracking experiments to compare results.
Data management: Understood how to collect, clean, and organize data.
Continual learning: Explored how models can keep improving with new data.
Model deployment: Learned ways to move models from notebooks to production.
Model monitoring: Focused on tracking performance after deployment.
Project management: Picked up how to plan and manage ML projects end-to-end.
I didn’t go deep like an ML engineer, but I wanted to know enough so I could speak the same language.
Link of the Resources:
3. Data Version Control (DVC)
One of my early frustrations was Why can’t I just use Git for datasets?
Well, because datasets are huge. That is when I discovered DVC (Data Version Control).
It was like a lightbulb moment. Just like Git manages code, DVC manages your datasets and models. I practiced by versioning CSV files and later full training datasets.
Link of the Resources:
DVG Docs
4. Experiment Tracking & Model Registry
I still remember training my first model and then forgetting what parameters I had used. Total Mess. That is when I learned about experiment tracking. Tools like MLflow let you log experiments, hyperparameters, and results. Suddenly, I was not juggling Excel sheets anymore and had a proper record of everything.
Link of the Resources:
MLOps ZoomCamp Youtube Playlist
5. CI + CD + CT
As an engineer, I was already comfortable with CI/CD. But MLOps introduces a new term called Continuous Training, or CT. I had seen models that worked perfectly for weeks but then started failing silently when new data arrived. Continuous Training ensures that the model keeps learning from fresh data regularly so it stays accurate and does not break over time. That is when I truly understood why CT is so important in MLOps.
Link of the Resources:
Before moving forward You can also follow a Detailed Video here:
6. Model Serving
Training a model is one thing, but making it available for real users or applications is a whole different challenge. Model serving means turning your trained model into an API or service that can respond to requests in real time. I struggled a lot here. My first FastAPI app crashed under load, and my first KServe attempt failed because I misconfigured Istio.
Link of the Resources:
7. GitHub Actions
GitHub Actions is great for setting up basic ML pipelines and automating smaller projects. I used it to automatically run training scripts, test models, and deploy simple APIs. For bigger projects with large datasets or more complex workflows, tools like Kubeflow are a better choice because they can handle heavy workloads and orchestrate the entire ML lifecycle.
Link of the Resources:
8. Monitoring & Observability
I quickly learned that models in production can lose accuracy over time, so monitoring is essential. Start with basic monitoring using tools like Prometheus, Grafana, and ELK to track metrics, logs, and system performance. Monitoring checks how your model performs and if it is running smoothly. Observability gives you the tools to understand what is happening behind the scenes.
Later, I used Evidently AI to watch for data changes and model drift, keeping everything reliable in production.
Link of the Resources:
9. Feature Stores
Feature stores are essential for managing features used in offline training and online serving. Feast is a popular option that works well with Redis or Dragonfly. I spent a good amount of time exploring their GitHub repos because they have excellent examples in Python and Node.js.
Link of the Resources:
Additional Resources I Loved
While the above formed the backbone of my learning, these resources helped me connect the dots and see the bigger MLOps picture:
Practitioner Guide to MLOps: A framework for continuous delivery and automation of machine learning: Great for understanding the workflow and best practices.
MLOps: Continuous delivery and automation pipelines in machine learning: Helps tie theory with practical pipelines.
My MLOps Roadmap MindMap
To make everything visual and easy to follow, I created a proper MLOps roadmap for you all and a dedicated video explaining each part of it.
I want to be honest, sharing these resources doesn’t mean I started from zero. I had prior DevOps and MLOps experience, but these resources helped me learn faster, structure my journey, and actually master MLOps.
Start small, stay curious, and get hands-on.
Remember, MLOps is a journey, not a checklist. Celebrate small wins, experiment often, and always keep exploring.
Until next time, keep building, keep experimenting, and keep exploring your AKVAverse. 💙
Abhishek Veeramalla, aka the AKVAman
Thanks for sharing each and every link😍. Really a person can learn end to end MlOps from this newsletter ✨
Thank you for sharing insight full information 👍