Core AI: Build AI
Eight modules take you from your first terminal command to a deployed ML system: Python and Git, maths, data engineering, machine learning, deep learning, NLP and LLMs, computer vision and MLOps. Every module ends in a graded build.
One level: Architect
Core AI is taught as a single course. There is no cut-down version: the whole program is the Architect level.
Architect level
From your first terminal command to a deployed ML system.
- Free preview first: Introduction and Preview
- 8 modules and 34 sections, about 156 to 189 hours
- A graded build at the end of every module
- Lifetime access, one payment
Already enrolled? Log in
Who it's for
- CS graduates and developers pivoting into AI/ML
- Engineers who want to understand models, not just call APIs
- Career-switchers willing to put in about 156 to 189 hours of focused effort
What you'll be able to do
- Explain and apply the maths behind ML with code
- Build data pipelines and features in Python and pandas
- Train, evaluate and select classical ML models
- Build neural networks by hand, then in PyTorch
- Work with tokenisation, embeddings, Transformers and LLM APIs; fine-tune models
- Build vision systems: detection, segmentation, OCR and multimodal
- Track experiments, serve models, monitor and ship with CI/CD
8 modules, 34 sections
MODULE 1Python, Terminal, Git and AI Tools
Never coded or never used Git? Start here. Experienced developers can skip it with a 15-minute self-check; it does not lock Module 2.
- Your machine, the terminal and environments
- Python from zero, for data work
- Git and GitHub, properly
- Working with AI tools: build, verify, do not trust
MODULE 2Mathematical Foundations
Ground truth for everything that follows, through visual, code-first exercises before formal notation.
- Linear algebra for ML
- Calculus and optimisation
- Probability and statistics
- Information theory and module project
MODULE 3Programming and Data Engineering
Get data in, make it trustworthy, turn it into features without leakage, and ship a pipeline someone else can re-run.
- Pandas for data work: streams, tables, joins
- Acquiring and cleaning messy data
- Features without leakage
- Storage, versioning and the reproducible pipeline
MODULE 4Machine Learning
Classical ML and the full train, evaluate and iterate cycle, compressed and hands-on.
- Regression: from the normal equation to regularised models
- Classification: logistic regression, trees, SVMs and boosting
- Unsupervised learning: clustering, anomalies and association rules
- Honest evaluation, tuning and the module project
Module project: Credit-risk scoring
MODULE 5Deep Learning
From a neuron and its gradient to Transformers: built by hand first, then in PyTorch, with tests.
- Neural network fundamentals: build one by hand
- Training deep networks with PyTorch
- CNNs and transfer learning
- Sequence models, Transformers and the module project
MODULE 6NLP and Large Language Models
From tokens and retrieval to tool-using agents, evaluation and security: build a language-model application you can test and defend.
- Text representations: from characters to contextual vectors
- LLM architecture, prompting and APIs
- Fine-tuning and adaptation: LoRA, data and preference tuning
- NLP applications and the module project
- Agents, tool use, MCP, evals and security
Module project: Multilingual support bot
MODULE 7Computer Vision
Image and video understanding: classical operations, detection and segmentation, transformers, generative models and OCR.
- Image fundamentals: pixels, classical vision and safe data pipelines
- Object detection and segmentation: boxes, metrics and masks
- Vision transformers, CLIP and generative vision
- Video, multimodal models, OCR and the shelf-analyser project
Module project: Shelf analyser
MODULE 8MLOps and Deployment
From notebook to production: one repository that tracks, serves, containerises, monitors and releases your credit-risk model.
- Tracked, reproducible training: start the capstone repository
- A tested service: contracts, no skew, safe logs, metrics and ONNX
- Containers, deployment and capacity: Docker, Kubernetes and cloud endpoints
- Monitoring, release engineering and the capstone build
- Architect deliverables: design, evaluate, govern
Module project: Capstone: the credit-risk model, tracked, served, containerised, monitored and released
Learn
Each section opens with its learning objectives and where the skill shows up in a 2026 job, then 12 or more teaching cards you open and study.
Try
"Try it yourself" prompts you paste into Claude or ChatGPT and runnable code with real output.
Prove
A 10-question quiz per section. Pass at 80% to unlock the next.
A graded build in every module
- Runnable code with real output in each section
- A project at the end of each of the 8 modules, graded against a written checklist
- A deployed ML system by the end of Module 8
Quizzes check recall; the module builds show the skill.
Sample build task
From the machine learning module: build a credit-risk scoring model for a personal-loan book (synthetic data) as one runnable script plus tests. You pick the model with cross-validation, tune it, and report calibration, a cost-based decision threshold chosen without touching the test set, and an error analysis. You also write a leakage canary test that proves your pipeline does not peek at held-out data, and a model card. You hand in the code, tests, evaluation report and model card, and they are scored against a published rubric.
Ready to start?
Not sure which track? The free 100-question assessment tells you.
