Practice · 01
Data
Collect, clean and label what the model will learn from.
Risk
Biased or leaked data bakes problems in from the start.
Make itthink.
AI is reshaping both sides of security and every kind of startup. We learn how models work well enough to build with them, and to question them.
Exp. 01Hypothesis
You don't need to be a researcher to understand AI. You need curiosity, some maths, and a place to try things.
We're building that place.
Note
Coming from security, we care as much about how AI fails as how it works.
Exp. 02Live experiment
Move, and watch it fire.
Run your cursor up and down the diagram. The input layer follows you, and the signal ripples forward through the network.
Fig. N1 — Forward pass4 · 6 · 6 · 3 neurons
Exp. 03Spec sheet
Eight things we test.
Each card is a topic we explore, from the foundations to the frontier, always with an eye on where it can go wrong.
MC-01
Machine Learning Foundations
Data, features, models and evaluation: how a machine actually learns from examples.
MC-02
Deep Learning
Neural networks from a single neuron to the architectures behind modern AI.
MC-03
LLMs & Prompting
How large language models work, where they shine, and where they quietly fail.
MC-04
Agents & Automation
Chaining models with tools to get real work done, reliably.
MC-05
Computer Vision
Teaching machines to see: classification, detection and what can go wrong.
MC-06
Data & MLOps
Getting models out of notebooks and into products that keep working.
MC-07
AI for Security
Anomaly detection, triage and threat analysis with machine learning.
MC-08
Securing AI
Prompt injection, data poisoning and model misuse, and how to defend against them.
Exp. 04Model lifecycle
Data to deployment, and the risks.
Practice · 02
Train
Fit a model to the data and tune it.
Risk
Overfitting: it memorises instead of learning.
Practice · 03
Evaluate
Test on data it has never seen.
Risk
A good score on the wrong test proves nothing.
Practice · 04
Deploy
Put it in front of real users.
Risk
Real inputs are messier, and sometimes adversarial.
Practice · 05
Monitor
Watch accuracy, drift and cost over time.
Risk
Silent decay: the world changes, the model doesn't.
Exp. 05Field plates
How machines learn.
The ideas behind modern AI, drawn live: networks, attention, clustering and the long walk down a loss curve.
Fig. 01 — AttentionAI
Fig. 02 — ClusteringAI
Fig. 03 — Gradient descentAI
Fig. 04 — AI for defenceDefense
Exp. 06Deep tracks
Three ways in.
Build with models, understand the maths, or keep AI safe. Most members end up doing a bit of all three.
Track 011
Build with models
From your first classifier to shipping an LLM-powered tool: the practical path to building with AI.
Track 022
Understand the maths
Linear algebra, probability and optimisation, taught so the intuition sticks before the equations.
Track 033
Keep it safe
Where AI meets our security roots: responsible use, evaluation, and defending models from attack.
Exp. 07Conclusion
Exp. 08Index