40-hour programmes
Ten curated slices of the same curriculum. Each programme's workload is summed module by module from the registry, and no module was created for them: all of them already exist in the curriculum and are the same objects the tracks use.
A programme is a 40-hour slice. For the complete programme of an area, with all four levels and the integrative project: Tracks
AI Fundamentals
40 hoursBeginner10 modules
8 of 10 modules with a published syllabus
0 of 10 modules recorded
Covers the entire level-100 shared trunk: what AI, ML and DL are; Python for AI; essential mathematics I; data collection and cleaning; basic statistics and probability; a first scikit-learn model; ethics, bias and safety; and Git/GitHub as a portfolio. It closes with the two orientation modules that open the other areas — visual intuition for neural networks, and the systems view of how AI becomes a product.
Data for AI
40 hoursBeginner9 modules
6 of 9 modules with a published syllabus
0 of 9 modules recorded
Treats data as the object of work before the model: Python, essential mathematics, collection and cleaning with Pandas, statistics and a first supervised model, then the two modules that introduce images and text as data (pixels, channels and OpenCV; tokens, cleaning and regex). It closes with evaluation and experimentation — overfitting, cross-validation, tracking — and with data pipelines, which is where the MLOps area begins. It shares five modules with AI Fundamentals; they are the same object, not a copy.
Prompt Engineering and AI APIs
40 hoursIntermediate8 modules
5 of 8 modules with a published syllabus
0 of 8 modules recorded
Focuses on building on top of third-party models rather than training your own: OpenAI/Anthropic/OSS APIs, function calling and cost; shipping the result as a service with FastAPI and Docker; and evaluating what you built with benchmarks, LLM evals, A/B testing and red teaming. The groundwork comes in through Python, text as data and ethics/bias; the agent orientation comes in through the two introductory Agentic AI modules, which draw the line between a chatbot and an agent.
Neural Networks and Deep Learning
40 hoursIntermediate7 modules
4 of 7 modules with a published syllabus
0 of 7 modules recorded
Goes from visual intuition to implementation: gradients and applied linear algebra, a neural network written from scratch with no framework, the same network rewritten in PyTorch with autograd and GPU, and the discipline of evaluation and experimentation. The two area modules add CNNs — shared with Computer Vision — and regularisation/normalisation.
Agentic AI: Building Agents
40 hoursIntermediate7 modules
1 of 7 modules with a published syllabus
0 of 7 modules recorded
Works through the operational definition of an agent — an LLM plus tools plus a loop —, the practical distinction between a chatbot and an agent, and then the three modules that make up the engineering of the subject: function calling and tool use, memory and state, and the orchestration frameworks. The AI APIs and prompt engineering module comes in as shared groundwork, and agent security (prompt injection, sandboxing) closes it, because an agent holding tools is an attack surface.
RAG and Semantic Search
40 hoursIntermediate7 modules
3 of 7 modules with a published syllabus
0 of 7 modules recorded
Covers the whole path from text to a grounded answer: text as data, embeddings and semantic search, end-to-end RAG — a module shared with Agentic AI — and classification/NER. The application layer comes in through AI APIs and prompt engineering, and the rigour through evaluating AI systems, without which a RAG system is not verifiable. Data collection and cleaning is included because the corpus is half the problem.
Applied Computer Vision
40 hoursIntermediate7 modules
2 of 7 modules with a published syllabus
0 of 7 modules recorded
Starts from images as data — pixels, channels, OpenCV — and climbs through CNNs (a module shared with Deep Learning), classification with augmentation and object detection with YOLO, ending in segmentation. PyTorch and data cleaning come in as groundwork shared with the trunk.
Multi-Agent Systems and MCP
40 hoursAdvanced5 modules
1 of 5 modules with a published syllabus
0 of 5 modules recorded
Brings together the four level-300 modules of the Agentic AI area — multi-agent systems and orchestration, planning/reflection/self-correction, MCP and integration with real systems, and agent security — with the module on evaluating AI systems. It presupposes the agent-building programme or equivalent.
LLMs: Fine-tuning and Alignment
40 hoursAdvanced5 modules
3 of 5 modules with a published syllabus
0 of 5 modules recorded
Concentrates the advanced core of the shared trunk — the Transformer architecture with attention implemented from scratch, fine-tuning and transfer learning with LoRA/QLoRA, and training optimisation with mixed precision and schedulers — and adds the two level-300 modules of the NLP area: fine-tuning LLMs in practice, and alignment (RLHF/DPO).
MLOps and AI in Production
40 hoursAdvanced5 modules
1 of 5 modules with a published syllabus
0 of 5 modules recorded
Deals with what happens after the model works: cloud for AI with GPU, storage and cost; observability and monitoring; serving at scale with batching, caching and autoscaling; training optimisation; and AI platform architecture.
Why the programmes overlap
The 10 programmes declare 400 hours in total, backed by 46 distinct modules adding up to 278 hours — a reuse factor of 1.44×. The overlap is deliberate, and it is what can be checked: the same module code, opened from different programmes, leads to the same object.
Not everything fits into 40 hours. State of the art, distributed training, interpretability, edge deployment and the five integrative projects exist only inside the complete tracks: Tracks