Course · 12 modules · 87 lessons
Data Analytics III: Machine Learning & AI Engineering
A 12-week hands-on course that takes you from "I can analyse data in pandas" to "I can build, evaluate and honestly report a predictive model and a grounded AI assistant." The first half is machine learning done properly: framing a prediction problem, choosing a baseline, regression and error metrics, gradient descent written from scratch, scikit-learn pipelines that never leak, classification with thresholds chosen by the real cost of mistakes, cross-validation, and a reproducible ML project evaluated once on a held-out test set. The second half is AI engineering: how language models work, prompts with output contracts validated in code, evaluation sets that catch regressions, retrieval with source citations that refuses to invent answers, and a read-only tool-using assistant with timeouts and failure handling. You finish with an integrated capstone — a real decision problem taken from question to data, dashboard or report, and a justified model or grounded assistant — defended in a 10-minute presentation. Entry requirement: Python functions, pandas, charts and basic statistics (Data Analytics II: Python for Data Analytics, or equivalent). Expect 10–15 hours per week of lessons, labs and a weekly project.
You can preview 3 lessons free.
Enroll for full access to every lesson, practical and instructor review.
01Week 1 — Prediction Problems & Data
- Using AI in this course: the rulesFree preview20m
- What machine learning is (and is not)Free preview50m
- Framing a prediction: decision, timing, cost50m
- Is this data fit for prediction?45m
- Leakage: when the answer sneaks into the question45m
- Lab: set up and inspect three datasetsPractical90m
- Lab: compute baselines by hand and in codePractical75m
- Weekly project: a written experiment planProjectFree preview180m
02Week 2 — Regression & Error
- Linear prediction: weights, bias and a line of best fit50m
- Residuals, MAE, MSE and RMSE50m
- Reading residuals and beating a baseline50m
- Many features: coefficients and their limits50m
- Lab: loss by hand, then in codePractical75m
- Lab: housing regression with residual plotsPractical90m
- Weekly project: regression experiment reportProject180m
03Week 3 — Training & Optimisation
- Loss and gradients: how a model learns50m
- Gradient descent from scratch in NumPy55m
- Learning rate, batch size and epochs50m
- Underfitting, overfitting and validation loss50m
- Lab: one parameter at a timePractical90m
- Lab: provoke and diagnose overfittingPractical75m
- Weekly project: a reproducible training experiment logProject180m
04Week 4 — Data Preparation & Pipelines
- Split first: train, validation and test50m
- Missing values, scaling and encoding50m
- Pipelines and ColumnTransformer55m
- Features, duplicates, outliers and imbalance50m
- Lab: watch leakage invent a modelPractical75m
- Lab: a full preprocessing pipeline, scaled vs unscaledPractical90m
- Weekly project: a leak-proof pipeline for your problemProject180m
- When AI suggests a feature that leaks30m
05Week 5 — Classification & Metrics
- Logistic regression: scores, sigmoid and thresholds50m
- Confusion matrix, accuracy, precision and recall55m
- ROC curves, AUC and precision-recall curves50m
- Choosing a threshold by the cost of mistakes50m
- Lab: a breast-cancer classifier with deliberate thresholdsPractical90m
- Lab: when 99% accuracy means nothingPractical75m
- Weekly project: classification reportProject180m
06Week 6 — Experiments & Model Selection
- Cross-validation and hyperparameters50m
- Searching hyperparameters inside a pipeline50m
- Error analysis, subgroups and uncertainty50m
- Documenting experiments and avoiding causal claims45m
- Lab: two approaches, one split, honest selectionPractical90m
- Lab: find where the model failsPractical75m
- Weekly project: experiment table and recommendationProject180m
- "The AI said my model is 99% accurate": checking claims25m
07Week 7 — Neural Networks, Embeddings & the ML Project
- Neural networks: layers of learned computations50m
- Reading training and validation curves45m
- Encodings and embeddings: sparse vs dense vectors50m
- Fine-tuning, distillation and scoping your ML project45m
- Lab: a neural network vs your best modelPractical90m
- Lab: similarity with sparse and dense vectorsPractical75m
- Machine learning project: build, evaluate and hand offProject300m
08Week 8 — Language Models, Prompts & Output Contracts
- How language models work, practically50m
- Hallucination, verification, cost and local vs hosted50m
- Calling a model from Python, provider-neutrally50m
- Prompts and output contracts55m
- Lab: the same questions with and without contextPractical90m
- Lab: a narrow extraction task with schema validationPractical90m
- Weekly project: context report and versioned promptProject180m
- Using model APIs responsibly: data, cost and NDPA25m
09Week 9 — Evaluating AI Systems
- Evaluation starts with success criteria50m
- Automated checks and human review50m
- Failure categories, regressions and trade-offs50m
- What an evaluation does and does not prove45m
- Lab: build a 20-case evaluation harnessPractical90m
- Lab: compare two prompt versions and review regressionsPractical90m
- Weekly project: an evaluation reportProject180m
10Week 10 — Retrieval & Grounded Answers
- Retrieval: answering from your documents50m
- Chunking and a TF-IDF index in code55m
- Grounded answers with citations50m
- Retrieval failure vs generation failure50m
- Lab: index a small authorised collectionPractical90m
- Lab: evaluate retrieval and answers separatelyPractical90m
- Weekly project: a grounded Q&A over your documentsProject180m
11Week 11 — Tools, Agents & the AI Project
- Tools and agents: letting a model ask for data55m
- The agent loop, validation and limits55m
- Application reliability: configuration, diagnostics, safe failure50m
- Scoping a bounded assistant45m
- Lab: a read-only tool, tested five waysPractical90m
- Lab: evaluate the agent end to endPractical90m
- AI engineering project: a bounded, evaluated assistantProject300m
12Week 12 — Integrated Capstone
- From question to recommendation: scoping the capstone50m
- Lab: build sprint with traceable metricsPractical120m
- Lab: 10-minute presentation and defence rehearsalPractical90m
- Career pathways: analyst, ML engineer, AI engineer40m
- Integrated capstone: decision problem to defended recommendationProject480m
- Defending AI-assisted work in your capstone20m
Already enrolled? Sign in with your enrollment email to unlock all lessons.