Us Oldies have all started with Andrew NG’s Machine learning and deep learning specialisation on Courser (youtube) but feel free to follow along the Andre Karpathy’s Zero to Hero series. This is a great code-first introduction for someone looking to jump directly into the world of LLM’s and skip the Classical Machine learning stuff (Linear Regression, SVM’s Decision tree’s etc)

The above two resources are a great place to get your hands dirtied but building your skills one brick at a time is how you actually fall in love with this field. Here’s how you do that

There are a few basic skills that everyone starting out in this field would need , they are

1. Coding : Python is the non negotiable language that you need to know in this field. Aside from basic syntax, it’s great to have knowledge about Object Oriented Programming concepts like Inheritance, Polymorphism, Abstraction. Writing your own magic methods, iterators familiarity with libraries like Numpy , Pandas, Matlplotlib and Pytorch is great to have. Remember that these skill develop over time and there is absolutely no need to follow a bottom up path of learning. Learn how to work with CSV’s , what is an array tensor etch. That is never think that only after learning everything about Numpy will you start your first ML project. Start with building a simple HTTP server in python, or if you want to jump into ML directly, build a model on the classic Titanic dataset by following along any of these notebooks on kaggle

2. Linear Algebra and Statistics :
1. Deep Learning Book: This book was written by the forefather’s of AI and part 1 of it is the least you’ll need to understand the Math’s that you will find in Research papers. I also use it as a refresher before interviews. Make sure to learn about Bayes theorem, Eigen values, Singular value decomposition, Gaussian Distributions in the least. After going through this make sure you are able to write down common loss functions, simple algorithms in mathematical notations
2. For a more detailed deep dive study Gilbert Stang’s lectures on Linear algebra and MIT open courseware course on statistics. There are a lot of great textbooks out there as well.

Done with the basics? It’s time to delve into something deeper. You can choose to start with either of the three fields. It’s important to keep on building

1. Classical Machine Learning : I love the book Hand’s on Machine Learning with SciKit Learn and Keras because it teaches all the basic concepts which I recommend everyone should learn regardless of the fact that they want to learn Vision or NLP in the future. Through this book I got a deeper understanding of building simple Linear and Logistic regression, Support vector machines, Binary, Multi-class, Multi-label loss functions, Train-Test-Validation split, Nearest Neighbour algorithm’s, Principal Component Analysis, Decision trees and Support Vector Machines. These are concepts that you should know even if you don’t go through this book and will keep on coming up in Deep Learning related concepts as well.

Don’t go through the “Keras” Part of this book as Keras and TensorFlow are not used in the community so much in 2025 (Unless you join Google, XD)

1. Computer Vision: CS -231n (Latest) is a gold mine to learn the basics of Computer Vision. Make sure to cover topics like Multi layer Perceptrons , Convolution operation, Evolution of architectures (AlexNet to Resnet to Vision Transformers) , Batch Normalisation, regularisation (Drop out), Optimizers, Augmentations for images, Attention framework. A good way to test yourself is to write your own CNN from Scratch in Python and achieve 99% accuracy on the MNIST dataset. Having a good grasp of different type of Convolution operations, skip connections, Global/Max/Average Pooling layers are signs you can call yourself a “Computer vision person”

Then Move on to Vision Transformers after learning the attention mechanism. Code out a Vision transformer from scratch and see if it able to beat CNN’s on any dataset . Try out Grad Cams to see exactly what these models look at when they decide if the image has a hotdog or a Dachshund.

By this time you should find yourself building toy projects or implementing research papers, downloading different datasets, delving into the official code of HuggingFace / PyTorch etc. Join an ongoing (or completed) competition on Kaggle and learn from other’s Notebooks and try to push yourself up the leaderboard by submitting solutions of your own.

Studying Classical Image Processing Algorithms is also a cherry on top (The old Professors in India would fall in Love with you if you do this). Jokes aside, Image processing algorithms teach you a lot about how to deal with Images. I found myself building great intuition about how to manipulate and analyse Image datasets. Things like Denoising, Contrast Enhancement, Image Histograms, HSV channels are important knowledge that help you later in the field, especially if you decide to purse Medical Applications of Computer Vision

2. Natural Language Processing: This is where you get the learn what the hype of LLM’s is about. Natural Language Processing is about dealing with text data , just like Computer Vision was about dealing with Image data.

CS-224n (Another stanford lecture series) is a good place to start. Make sure to learn about different ways of Text Preprocessing (Normalisation, stemming , lemmatisation, Unicode characters and representation of foreign language characters) , RNN’s and LSTM’s (outdated so no need to delve deeper) , Attention and Transformer Mechanisms (Encoder and Decoders;BERT, GPT), Common NLP tasks like Sentiment Analysis and Named Entity Recognition, Machine translation etc.

General Advice

• Keep Reading and Implementing (and hopefully Write!) research papers.

• Join Kaggle competitions :/

• Time series, Reinforcement Learning, Recommender systems, are all different fields that can be explored. Explore these topics only to either prepare for an interview with a company/Prof or if you have already gotten a project in these fields.

• MLOP’s : Explore different techniques to deploy your ML models, can look into TorchServe, VLLM’s and AWS SageMaker Endpoints, hugging face spaces. It will be a huge plus point if the HR folks can directly click a link and see what your ML model is all about.

• Read and Learn from documentations! : Documentations should be your main source of knowledge about any language or framework. If you are in the habit of learning from “youtube videos” break that habit as soon as possible.

• Proficiency in PyTorch, Numpy and pandas. These are the 3 frameworks which you will have to learn better than English itself. Be generally aware of common functions that you use to manipulate tensors, data-frames etc and ensure that your are not reliant on Copilot or Cursor for all your needs. I learn PyTorch by reading the book : Deep Learning with Pytorch .

Note that PDF’s of most books mentioned here will be available free of charge on pirated website like Z-Lib or Anna’s archive.

There are a lot of sources on which quality content is available, therefore not adding anything except the resources which i already tagged. Go to Reddit / Twitter or to perplexity or Gemini Deep research about your specific needs !