Artificial Intelligence for Programmers Course

Lesson 1 • What AI Actually Is

Defines AI, machine learning, and deep learning as distinct layers. Grounds abstract definitions in concrete programming tasks developers already recognise.

Lesson 2 • History and Milestones of AI

Traces key breakthroughs from early perceptrons to transformer models. Provides context for why modern techniques dominate current practice.

Lesson 3 • Maths Essentials for AI

Reviews linear algebra, probability, and calculus concepts used daily in AI code. Focuses on intuition and code-level application rather than formal proofs.

Lesson 4 • Core AI Problem Categories

Maps supervised, unsupervised, and reinforcement learning to real engineering tasks. Enables developers to frame new problems within the correct learning paradigm.

Lesson 1 • Feature Engineering and Transformation

Transforms raw fields into model-ready features through encoding, scaling, and aggregation. Demonstrates how feature quality directly impacts model accuracy.

Lesson 2 • Data Collection and Ingestion

Covers web scraping, API ingestion, and database extraction for AI datasets. Establishes legal and ethical boundaries around data sourcing.

Lesson 3 • Building Reproducible Data Pipelines

Packages data transformations into versioned, testable pipeline stages. Introduces data versioning tools to track dataset changes across experiments.

Lesson 4 • Data Cleaning and Validation

Applies programmatic techniques to detect and fix missing values, outliers, and duplicates. Validates schemas and distributions before data enters a training pipeline.

Lesson 1 • Clustering and Dimensionality Reduction

Applies k-means, DBSCAN, PCA, and t-SNE to unsupervised discovery tasks. Teaches evaluation metrics for unlabelled data scenarios.

Lesson 2 • Tree-Based Models

Builds decision trees, random forests, and gradient-boosted trees with tunable hyperparameters. Explains ensemble mechanics that make boosting state-of-the-art on tabular data.

Lesson 3 • Linear and Logistic Regression

Implements gradient-descent regression from scratch before using library versions. Establishes the loss-optimisation loop that all subsequent models share.

Lesson 4 • Model Evaluation and Selection

Applies cross-validation, confusion matrices, ROC curves, and calibration plots to compare models. Prevents data leakage through correct train-validation-test splits.

Lesson 1 • Anatomy of a Neural Network

Defines layers, neurons, weights, biases, and activation functions as composable code objects. Connects the forward pass to a differentiable computation graph.

Lesson 2 • Training at Scale with Frameworks

Implements training loops in PyTorch and TensorFlow with GPU acceleration. Covers data loaders, mixed-precision training, and checkpoint management.

Lesson 3 • Backpropagation and Optimisers

Derives backpropagation via the chain rule and implements it manually before using autograd. Compares SGD, Adam, and RMSProp on convergence speed and stability.

Lesson 4 • Debugging Neural Network Training

Diagnoses vanishing gradients, exploding gradients, and dead neurons using diagnostic tools. Establishes a systematic debugging workflow applicable to any architecture.

Lesson 1 • Transfer Learning and Fine-Tuning

Adapts pretrained ImageNet models to new tasks with limited labelled data. Covers layer freezing strategies and learning rate scheduling for fine-tuning.

Lesson 2 • Object Detection Architectures

Implements anchor-based and anchor-free detectors for locating objects in images. Evaluates detectors using mean average precision and IoU thresholds.

Lesson 3 • Convolutional Neural Networks

Explains convolution, pooling, and receptive fields as spatial feature extractors. Implements a CNN from scratch and traces feature maps through each layer.

Lesson 4 • Vision Data Augmentation

Designs augmentation pipelines that improve generalisation without corrupting labels. Covers geometric, colour, and advanced augmentation strategies.

Lesson 1 • Transformer Architecture Deep Dive

Implements multi-head self-attention and positional encoding from scratch. Traces how BERT and GPT differ in pretraining objectives and downstream use.

Lesson 2 • Fine-Tuning Pretrained Language Models

Adapts BERT-family models to classification, NER, and QA tasks with task-specific heads. Applies parameter-efficient methods like LoRA to reduce compute cost.

Lesson 3 • Prompt Engineering and LLM APIs

Designs effective prompts for instruction-tuned LLMs using zero-shot and few-shot patterns. Integrates LLM APIs into production code with structured output parsing.

Lesson 4 • Text Representation and Preprocessing

Converts raw text to numerical representations via tokenisation, embeddings, and TF-IDF. Establishes the preprocessing chain that feeds any NLP model.

Lesson 1 • Model Packaging and Containerisation

Serialises models and wraps them in Docker containers with dependency isolation. Prepares portable model images deployable to any cloud or on-premise target.

Lesson 2 • CI/CD for Machine Learning

Automates model testing, validation, and deployment through ML-specific CI/CD pipelines. Integrates data and model quality gates before any production promotion.

Lesson 3 • Experiment Tracking and Reproducibility

Logs hyperparameters, metrics, and artefacts with MLflow and Weights and Biases. Enables reproducible experiment comparison across team members and runs.

Lesson 4 • Serving Models as APIs

Builds REST and gRPC inference endpoints with FastAPI and TorchServe. Applies batching, caching, and async handling to meet latency requirements.

Lesson 1 • Designing Multi-Model Pipelines

Composes specialised models into orchestrated pipelines with routing and fallback logic. Applies system design principles to AI-specific reliability requirements.

Lesson 2 • AI Agents and Tool Use

Builds LLM-powered agents that plan, call tools, and iterate toward goals autonomously. Implements ReAct and function-calling patterns with safety guardrails.

Lesson 3 • Responsible AI and Governance

Embeds fairness audits, explainability tools, and privacy-preserving techniques into system design. Aligns AI systems with organisational risk and compliance requirements.

Lesson 4 • Scalable AI Infrastructure

Provisions GPU clusters, distributed training jobs, and inference autoscaling on cloud platforms. Optimises infrastructure cost through spot instances and model quantisation.