AI/ML Fundamentals
The math you actually need, the models companies fine-tune, and the evaluation that stops you shipping a confident wrong answer.
Mastery unlocks after a 90% Quiz score
Real-world analogy
A model is a very expensive intern. It is fast, it is confident, and it will fabricate a citation unless you check its work.
◎ Why it matters in production
Job posts now say 'familiar with LLMs'. That does not mean you used ChatGPT. It means you can talk about eval, retrieval, and failure modes.
Fundamental concepts you will master
Fit vs. memorise
Train, val, test. Leakage is how you get a 99% accuracy that dies in production.
Loss is a compass
Gradient descent is just walking downhill. The interesting part is the shape of the hill.
Eval before demo
Precision, recall, calibration — pick the metric the product actually cares about.
LLMs as APIs
Context windows, tokens, temperature, and why the prompt is not the product.
Step-by-step curriculum
The supervised loop
Features, labels, split, fit, evaluate. Do it by hand once.
Classical models
Linear, trees, boosting — still what a lot of tabular jobs run.
Neural nets
Layers, activations, backprop at the level you can explain on a whiteboard.
LLMs
Tokenisation, decoding, fine-tune vs RAG vs agents.
Eval & safety
Hallucinations, bias, and the test set you should have held out.
Ready to see it in action?
Build your own path, step through operations, and watch mastery unlock after 90%.
Pointer Phantom
Direct Byte Access
+1,250 XP → Cache Ghost