Modern AI Systems and Methods

A field guide to how modern AI actually fits together

Nineteen chapters mapping supervised learning through agents, evaluation, and responsible AI — real computed visualizations, not diagrams, and every chapter linked to the ones around it.

Modern AI is not one ladder of methods — it’s a set of overlapping problem framings that a real system usually combines. A search assistant might use retrieval, ranking, embeddings, an LLM, calibration, experimentation, privacy constraints, and production monitoring, all at once. This is a field guide to those framings: what each one is, what problem it actually solves, what it assumes, how it connects to its neighbors, and where it shows up in systems you’d recognize.

Three commitments shape everything here:

  1. See it, don’t just read about it. Nearly every chapter includes at least one real, computed visualization — a fitted decision boundary, an MCMC trace, a value-iteration heatmap, a diffusion process actually run forward and backward — generated from a small script on toy or synthetic data. Not a decorative diagram standing in for the idea.
  2. Explain the mechanism, not just the name. Every chapter tries to answer why a method works the way it does, not just list that it exists. Bullet-point glossaries are easy to skim and easy to forget; a worked explanation sticks.
  3. Map the connections. Almost nothing here is an island. Regularization is a Bayesian prior in disguise. A recommender’s matrix factorization is a latent-variable model. RAG is retrieval wearing a generative model’s clothes. Each chapter ends by naming its neighbors explicitly, and the map draws the whole graph at once.

0.1 How this site is organized

Nineteen chapters — the map plus the eighteen framings it maps — grouped into seven parts that follow one narrative arc: predicting from labeled data, modeling uncertainty and hidden structure, learning representations, making decisions and generating content, knowing whether any of it actually works, and operating it reliably and responsibly in the world. Start at the map for the whole landscape in one page, then read chapters in order for broad coverage, or jump straight to whatever you’re trying to understand.

0.2 Reading this site vs. reading the source

Every figure here was generated by a short, readable Python script in scripts/figures/ on the project’s GitHub repo — real numpy/scikit-learn computations on toy data, not hand-drawn. If a figure raises a question the prose doesn’t answer, the script that made it usually will.