Data & Machine Learning Foundations
Six weeks on what machine learning actually does, so you can reason about model behaviour rather than treat it as magic. Enough theory to make good engineering decisions, and no more.
Join the waitlistWho should take this program?
You'll get the most from this if
- Comfortable writing Python
- School-level mathematics
This is not for
- Anyone seeking a research or PhD-track ML course
- Anyone who wants to train frontier models from scratch
What you will build
- A trained and evaluated classifier with an honest error analysis
- A feature pipeline that handles messy real data
- A short written assessment of where your model fails and why
What you learn, week by week
Module 1 — What ML can and cannot do
- Frame a problem as a learning task, or reject it
- Recognise problems better solved with rules
- Set a baseline before touching a model
Module 2 — Data before models
- Find leakage before it flatters your results
- Handle missing and imbalanced data honestly
- Build a reproducible split
Module 3 — Training and tuning
- Train a model without fooling yourself
- Tune without overfitting the validation set
- Know when more data beats a better model
Module 4 — Evaluation that means something
- Choose metrics that match the business cost
- Read a confusion matrix properly
- Report uncertainty
Module 5 — Embeddings
- Understand what an embedding represents
- Use vector similarity correctly
- See where semantic search fails
Module 6 — Error analysis
- Categorise failures instead of counting them
- Decide what to fix first
- Write the analysis someone else can act on
How this program handles evaluation and governance
Most AI harm starts in the data, not the model. This course spends real time on leakage, imbalance, and evaluation metrics that flatter a system while hiding who it fails. You will write an error analysis that names which groups or cases your model gets wrong — the habit every governance framework later depends on.
Tools and stack you will use
Fees
Request pricing
Fees depend on cohort, format, and whether this runs for an individual or a team. We'll send the full breakdown.
Get fee detailsQuestions people ask
How much maths?
Enough to reason about behaviour. We do not derive gradients.
Will I train large language models?
No. You will understand how models learn, which is what you need to work with LLMs sensibly.
Is this needed before the agentic track?
Not strictly, but people who skip it tend to misdiagnose model failures as prompt problems.