How to use this book

The purpose of computing is insight, not numbers.

R. W. Hamming, Numerical Methods for Scientists and Engineers (1962)

Before the orientation, a word on why this book exists and why its subject belongs inside your training rather than at its edge. If you are a first-year medical or public-health student holding this book, the working life you are preparing for will not resemble the one your instructors trained in. Within the span of a single graduate program, generative AI has moved from a curiosity to a fixture of the clinic and the health department. Models now conduct diagnostic dialogue that clinicians rate favorably against their own (Tu et al., 2025), and the field has begun, in earnest, to build generalist medical systems that read images, notes, and signals within one model (Moor et al., 2023). The preface makes the fuller case; here we state only its conclusion. The professionals who will do the best work with these systems are not those who trust them most, nor those who refuse them, but those who have learned to direct them and to verify what comes back. That is a learnable skill, and it is the subject of this book.

We would enter one caution against waiting. It is tempting to treat a fast-moving technology as something to take up later, once it has settled. It will not settle on any schedule that rewards waiting: across the period in which this book was written, the capability frontier advanced by roughly a model generation every few months. The reader who learns the method now, while the tools are still imperfect and their failures are still instructive, will be better prepared than the reader who waits for a stability that is not coming.

We begin with a short orientation chapter, in which we set out the chapter template, the cross-references, and the conventions used throughout the book. It is, in general, the same template inherited from the sister volumes, with two adjustments specific to this book.

Map of the book

We begin with the shape of the whole book. Figure 1 lays out the five parts and their fourteen chapters, together with the two reading dependencies the preface notes: Chapter 8 builds on Chapter 3, and Chapter 14 builds on Chapter 13. A reader who wants a path through the book, rather than a straight march front to back, can read it off the figure directly, since most chapters can be entered on their own and only the two dashed arrows impose an order.

flowchart TB
  subgraph P1 ["Part I. Foundations"]
    direction LR
    C1["Ch 1<br/>Genesis of<br/>generative AI"] --> C2["Ch 2<br/>Landscape and<br/>capability classes"] --> C3["Ch 3<br/>Reasoning, context,<br/>verification"] --> C3B["Ch 4<br/>Programming<br/>with AI"]
  end
  subgraph P2 ["Part II. Grounding models in evidence and data"]
    direction LR
    C4["Ch 5<br/>RAG over<br/>biomedical corpora"] --> C5["Ch 6<br/>Synthetic data<br/>and privacy"] --> C6["Ch 7<br/>Multimodal<br/>medical AI"]
  end
  subgraph P3 ["Part III. Agentic workflows and tool use"]
    direction LR
    C7["Ch 8<br/>Agents, tools,<br/>and MCP"] --> C8["Ch 9<br/>Deep research and<br/>evidence synthesis"]
  end
  subgraph P4 ["Part IV. Evaluation, safety, and governance"]
    direction LR
    C9["Ch 10<br/>Evaluation beyond<br/>the benchmark"] --> C10["Ch 11<br/>Safety, bias,<br/>red-teaming"] --> C11["Ch 12<br/>Regulation, privacy,<br/>the IRB"]
  end
  subgraph P5 ["Part V. Customisation, deployment, and practice"]
    direction LR
    C12["Ch 13<br/>Fine-tuning and<br/>AI-augmented teams"] --> C13["Ch 14<br/>Deployment in<br/>practice"]
  end
  P1 --> P2 --> P3 --> P4 --> P5
  C7 -.->|"builds on"| C3
  C13 -.->|"builds on"| C12
Figure 1: A map of the book: five parts and fourteen chapters. Solid arrows follow the default reading order within and across parts; the two dashed arrows mark the dependencies noted in the preface, where Chapter 8 builds on Chapter 3 and Chapter 14 on Chapter 13. Most other chapters may be entered directly.

A word on emphasis. The book serves the full range of health-sciences researchers, and its worked examples are drawn from across that range (clinical, pharmacological, laboratory). The center of gravity, however, sits with public health and population-scale work: disease surveillance, epidemiology, health equity and disparities, and evidence for policy. A reader in a School of Public Health should, in general, find the running examples pitched at the settings in which they already work; a reader elsewhere in the health sciences will, we hope, find the transfer a short one.

Chapter template

Every content chapter shall follow the same nine-section structure:

  1. Learning objectives. A bulleted list of capabilities the reader should have once they have worked through the chapter.
  2. Orientation. Two to four paragraphs of prose framing: what the chapter is, why it matters, and how it relates to adjacent chapters.
  3. The researcher’s contribution. A front-loaded section articulating the judgments at the center of the chapter that no large language model can make on the reader’s behalf. Three to five paragraphs, each leading with a bold-tag judgment and followed by elaboration.
  4. Content sections. The chapter’s substantive material, broken into sections with descriptive headings. Collapsible Check your understanding callouts appear at natural pauses.
  5. Worked example. A concrete worked example, ideally biomedical, that exercises the chapter’s tools end to end.
  6. Collaborating with an LLM on topic. Three prompt patterns paired with what to watch for and how to verify, specific to the chapter’s content.
  7. Principle in use. Three habits that define defensible work in this area.
  8. Exercises. Five exercises ranging from short conceptual checks to extended applied work.
  9. Further reading. Canonical, modern applied, and package-documentation pointers.

Table 1 distils the same nine sections into a one-line statement of what each is for, as a quick reference to return to while navigating any chapter.

Table 1: A quick-reference legend for the nine-section chapter template.
Section What it is for
Learning objectives The capabilities to acquire; use them to self-assess afterwards
Orientation Why the topic matters and how it joins the chapters around it
The researcher’s contribution The judgments no model can make on the reader’s behalf
Content sections The substantive method, with Check your understanding pauses
Worked example The chapter’s tools exercised end to end on one problem
Collaborating with an LLM Prompt patterns paired with what to watch for and how to verify
Principle in use The habits that make work in the area defensible
Exercises Graded practice, which this volume treats as its assessment
Further reading Where to go next, from canonical to current

Adjustments specific to this book

Three adjustments distinguish this volume from the introductory and advanced sister volumes:

  • No prerequisites quiz. The introductory volume opens each chapter with a pre-flight quiz to surface missing prerequisites; this volume, on the other hand, assumes the prerequisite material is in place and so skips the quiz.
  • No practice test. The introductory volume closes with a self-administered practice test; this volume, by contrast, treats the exercises as the equivalent assessment.
  • Researcher-centerd framing. The sister volumes address the biostatistician explicitly. This volume, whose audience extends across health-sciences research (clinicians, biostatisticians, epidemiologists, pharmacists, public-health researchers), refers instead to the researcher: that is, whoever is running the analysis and signing the report. Where a chapter contains a worked example with an explicit role, we name the role (clinical fellow, pharmacoepidemiologist, biostatistician); the surrounding prose, however, stays role-neutral.

Cross-references

References to chapters in this volume use @sec- labels: for example, @sec-agents-mcp resolves to Chapter 8. References to chapters in the sister volumes, on the other hand, use full URLs, since the volumes deploy as separate Quarto sites: SCAI lives at https://scai.rgtlab.org, the advanced volume at https://scai-advanced.rgtlab.org, and the practicum at https://practicum.rgtlab.org.

Conventions

We describe the visual cues used throughout the book on the Conventions page. Code blocks default to R; Python and shell snippets are labeled. LLM prompts are shown in fenced blocks with the prompt text and, where relevant, the literal verbatim model response.

Companion repositories

The book repository at https://github.com/rgt47/applied-genai contains:

  • docs/syllabi-survey.md, the curriculum survey that shaped the chapter list.
  • docs/mollick-digest.md, the 24-month One Useful Thing digest used to test the TOC for contemporariness.
  • references.bib, the working bibliography.

Both companion documents are intended as audit-trail artifacts; that is, they explain why the book is shaped the way it is, and we shall update them as the field evolves.