AI and machine learning

Time

35–44 weeks + Ongoing

Before this

Foundations of computer science

Leads to

Machine learning engineer, Data scientist, AI researcher, MLOps engineer

People comfortable with programming who enjoy math and experiments.

1. Stage 1: Math for machine learning , 6–8 weeks

   What to learn

   • Vectors and matrices
   • Derivatives and gradients
   • Probability
   • Statistics
   • Optimisation

   Free resources

   • Essence of Linear Algebra Video, 3Blue1Brown
   • Mathematics for Machine Learning Book, Deisenroth, Faisal and Ong (Cambridge University Press)
   • Statistics and probability Course, Khan Academy

   Build: Implement linear regression with gradient descent using only NumPy, and plot the loss going down.

   Move on when: You can explain gradient descent on a whiteboard.

2. Stage 2: Python for data , 3–4 weeks

   What to learn

   • NumPy arrays
   • pandas data frames
   • Cleaning messy data
   • Plotting
   • Jupyter notebooks

   Free resources

   • NumPy quickstart Docs, NumPy
   • pandas getting started Docs, pandas
   • Kaggle Learn Practice, Kaggle

   Build: Clean and explore a public dataset, and write down three things you found.

   Move on when: You can load, clean and summarise a dataset you have never seen.

3. Stage 3: Classical machine learning , 8–10 weeks

   What to learn

   • Training, validation and test sets
   • Overfitting and regularisation
   • Regression and classification
   • Decision trees and ensembles
   • Choosing the right metric

   Free resources

   • Machine Learning Crash Course Course, Google
   • An Introduction to Statistical Learning Book, James, Witten, Hastie, Tibshirani and Taylor
   • scikit-learn user guide Docs, scikit-learn

   Build: A model trained end to end, tested on data it never saw, with an honest note on where it fails.

   Move on when: You can tell when a good score is really data leakage.

4. Stage 4: Deep learning , 10–12 weeks

   What to learn

   • Neural networks and backpropagation
   • PyTorch
   • Convolutional networks
   • Transformers and attention
   • Training on a GPU

   Free resources

   • Practical Deep Learning for Coders Course, fast.ai
   • Neural Networks: Zero to Hero Video, Andrej Karpathy
   • Dive into Deep Learning Book, d2l.ai (open source)
   • PyTorch tutorials Docs, PyTorch Foundation

   Build: Train an image or text classifier and compare it with the classical model from the last stage.

   Move on when: You can build and debug a training loop yourself.

5. Stage 5: Language models , 8–10 weeks

   What to learn

   • Tokens and embeddings
   • How large language models are trained
   • Fine-tuning
   • Retrieval-augmented generation
   • Evaluating outputs

   Free resources

   • LLM Course Course, Hugging Face
   • CS224N: NLP with Deep Learning Course, Stanford University

   Build: A question-answering app over your own notes, with a small test set to measure how often it is right.

   Move on when: You can explain why a model gave a wrong answer and measure whether a change helped.

6. Stage 6: Put models into use, responsibly , Ongoing

   What to learn

   • Serving a model behind an API
   • Monitoring and data drift
   • Bias and fairness
   • Privacy
   • Attacks on AI systems

   Free resources

   • Made With ML Course, Goku Mohandas
   • Rules of Machine Learning Docs, Google
   • AI Risk Management Framework Docs, NIST

   Build: Deploy one of your models with monitoring and a short model card that says what it should not be used for.

   Move on when: This stage never ends. Keep shipping and measuring.

Certification roadmaps Back to all paths