teacher
Sign in
Track

AI Engineering

The math, models and engineering every AI engineer needs — from gradient descent to RAG and agents.

32 lessons~18 hours of lessonsTarget: Intermediate — build, evaluate and ship LLM-powered featuresPython
Reality check
  • Tools change every few months; the fundamentals (linear algebra, probability, optimisation, evaluation) do not. Invest there first.
  • Most real AI engineering is data quality, evaluation and plumbing — not training giant models.
  • You will need to read papers and documentation. These lessons give you the map, not every detail.
  • A demo that works once is not a product. Measure with evals before you trust any change.
  1. Stage 10/4

    Foundations

    Python, NumPy and just enough math and probability.

    1. 01Start HereWhat AI engineering actually involves, what this track covers, and the honest ratio of glamour to plumbing.15 min
    2. 02Python and NumPyArrays, shapes and broadcasting, the habits that make ML code short and the bugs that make it confusing.30 min
    3. 03Math for MLVectors, dot products, cosine similarity, derivatives and gradients, only the parts the rest of the track uses.35 min
    4. 04Probability and StatisticsProbability, expectation, Bayes' rule, distributions and sampling, the vocabulary every model evaluation uses.35 min
  2. Stage 20/7

    Machine learning

    Regression, classification, evaluation, overfitting and classic models.

    1. 01What Is Machine LearningLearning from examples instead of rules, the main kinds of ML, and the train, validate, test habit that keeps you honest.25 min
    2. 02Linear RegressionFit a straight line to data by minimising squared error, with gradient descent and with a closed-form solution.35 min
    3. 03Logistic RegressionTurn a weighted sum into a probability with the sigmoid, train it with log loss, and choose a decision threshold.35 min
    4. 04Evaluation MetricsAccuracy, precision, recall and F1 from the confusion matrix, the effect of class imbalance, and how to pick a metric.35 min
    5. 05Overfitting and RegularizationWhy a model that scores perfectly on its training data can fail on new data, and the standard ways to prevent it.30 min
    6. 06Trees and EnsemblesDecision trees split data with simple questions; random forests and boosting combine many trees into the strongest classic models for tabular data.35 min
    7. 07Unsupervised LearningFind structure in data that has no labels, using k-means clustering and PCA for dimensionality reduction.30 min
  3. Stage 30/6

    Deep learning

    Neural networks, backpropagation, training and PyTorch.

    1. 01Neural NetworksA neural network is layers of simple weighted sums followed by nonlinear functions, and stacking them lets it learn curved patterns.30 min
    2. 02BackpropagationHow a network works out, for every weight, which way to nudge it, using the chain rule applied backwards through a computation graph.35 min
    3. 03Training Deep NetworksThe practical knobs that decide whether a network trains well - batches, learning rate, optimizers, initialisation, normalisation - and how to debug a run that does not.35 min
    4. 04PyTorch BasicsTensors, automatic gradients, nn.Module and the standard training loop in PyTorch, the framework most deep learning code is written in.35 min
    5. 05Convolutional Neural NetworksHow small sliding filters let a network find patterns in images with far fewer weights than a fully connected layer.35 min
    6. 06EmbeddingsTurn words, images or any item into a vector of numbers so that similar things land close together, then compare them with cosine similarity.30 min
  4. Stage 40/6

    Transformers & LLMs

    Attention, the transformer, tokenization and how LLMs are trained and sampled.

    1. 01AttentionHow each token in a sequence looks at every other token and mixes in the information it needs, using queries, keys and values.35 min
    2. 02The Transformer ArchitectureHow attention, a small neural network, residual connections and normalisation stack into the block that every modern LLM repeats dozens of times.35 min
    3. 03TokenizationHow text becomes the integer ids a language model reads, why subword pieces won, and the odd behaviour that comes from it.30 min
    4. 04How LLMs Are TrainedNext-token prediction at scale, then instruction tuning and preference training, with the loss, the compute and the limits spelled out.35 min
    5. 05Decoding and SamplingHow a model's next-token probabilities become actual text, and what temperature, top-k and top-p really change.30 min
    6. 06Prompt EngineeringWrite prompts that are specific, structured and testable, and learn which tricks actually help and which are folklore.30 min
  5. Stage 50/9

    Building AI apps

    APIs, RAG, vector search, agents, evals, fine-tuning, serving and safety.

    1. 01LLM APIs and Structured OutputCall a hosted model from code, handle failures and cost, and turn free-form text into validated data your program can trust.35 min
    2. 02Retrieval-Augmented Generation (RAG)Give a model the right pages at question time by searching your documents first, then answering from what was found.40 min
    3. 03Vector Databases and Approximate SearchFind the nearest vectors among millions without comparing against all of them, using IVF and HNSW indexes and their speed versus recall trade-off.40 min
    4. 04Agents and Tool UseLet a model call functions in a loop to look things up and take actions, and understand MCP, the open protocol that standardises how tools plug in.40 min
    5. 05EvalsMeasure whether an LLM feature works with a test set, code-based checks and calibrated judges, so changes are decided by numbers and not by impressions.40 min
    6. 06Fine-Tuning and LoRADecide when training a model beats prompting or retrieval, and see how LoRA adapts a large model by training a few million extra parameters.40 min
    7. 07Serving and CostUnderstand where LLM latency and money go (prefill, decode, KV cache, batching) and the levers that cut both, with back-of-the-envelope maths.40 min
    8. 08Safety and GuardrailsLearn the main ways LLM applications get attacked or misbehave, and the layered controls (least privilege, validation, confirmation, testing) that contain the damage.35 min
    9. 09Multimodal ModelsSee how images become tokens for a transformer (ViT), how CLIP aligns pictures and text, and how vision-language models combine them with an LLM.40 min