Foreword: About the Author and the AI Office at KHCC

Before we open the first terminal, it’s worth saying who wrote this book and why it exists.

About Dr. Iyad Sultan

Iyad Sultan is a pediatric oncologist at King Hussein Cancer Center (KHCC) in Amman, Jordan, where he also serves as the Medical Director of the Artificial Intelligence Office. His clinical work centers on pediatric solid tumors — Wilms tumor, neuroblastoma, soft-tissue sarcomas — and his research, published in The Lancet Oncology, Journal of Clinical Oncology, Cancer, and Pediatric Blood & Cancer, has used large national datasets (notably the U.S. SEER registry) to ask questions about incidence, treatment patterns, and outcomes in childhood cancer that single-institution series cannot answer.

Several years ago he started building. What began as a few Python scripts for cleaning pathology reports has become an institutional program. Today the AI Office at KHCC ships pipelines that run every night against the hospital’s data, flagging clinical events, extracting structure from free text, and giving doctors and pharmacists tools they didn’t have before. He still rounds. He still writes the prompts. The point of this book is to take what he has learned standing between the clinical floor and the data team, and pass it to the next clinician who wants to learn the same skill.

About KHCC

King Hussein Cancer Center is the largest comprehensive cancer center in the Middle East. It treats roughly twelve thousand new cancer patients per year, runs a high-volume bone marrow transplant program (about a hundred and eighty allogeneic transplants a year, with a high proportion of haploidentical donors), and serves both adults and children from across the region. KHCC is JCI-accredited, ISO-certified, and is the only institution in the region with both a cancer center and a research arm under one roof. Its electronic medical record is built on VistA, a system originally developed by the U.S. Department of Veterans Affairs and substantially extended for cancer-specific workflows at KHCC.

About the AI Office

The AI Office at KHCC sits inside the King Hussein Cancer Center Institute (KHCCI), the research arm of the hospital. Its mandate is to put working AI into the hands of clinicians, not to write papers about it.

🔧 A note on naming. Through 2025 the office was called the AIDI Department (short for Artificial Intelligence Department). It was renamed the AI Office in 2026. You will still see the old short form scattered through the source code — the Databricks catalog is named aidi_catalog, repositories carry names like aidi-extractions, and a number of pipelines reference aidi/eval/run_deceased_cohort.py in their imports. These are real technical identifiers that downstream pipelines depend on, so they have not been renamed. When you see “aidi” in lowercase as part of a code identifier, it’s the historical name baked into the system. When you see “AI Office” in prose, it’s the office today. They are the same place.

The office is small by design. A handful of clinicians who code (Dr. Sultan and a rotating set of fellows), a handful of engineers and data scientists, a Databricks workspace whose data catalog is named aidi_catalog, and a strict rule that every pipeline that touches patient data is gated by an evaluation suite frozen against a cohort of deceased patients. No prompt change ships without an eval run. No extraction logic change ships without an eval run. The eval is the test.

What the office actually builds — the pipelines you’ll see referenced throughout this book — are described in Chapter 0.5. Read them. They are not props. They are real systems running on real KHCC data, and the techniques in this book are the techniques used to build them.

Why this book

For the first ten years of his career, Iyad shipped code by writing it himself, in Jupyter notebooks, late at night, after his pediatric tumor board day was over. That was sustainable. Then Claude Code arrived. Within a year, the rhythm of his work changed completely. The bottleneck was no longer how fast he could type Python. It was how clearly he could describe what he wanted, and how much trust he had built up in the verification systems (the eval cohort, the hooks, the deny lists) that would catch the mistakes.

That shift is available to any clinician with curiosity. It does not require a computer science degree. It requires understanding what a terminal is (Chapter 0), what an agent is (Chapter 1), what CLAUDE.md does (Chapter 10), and a small handful of habits about safety and verification that the book teaches by repetition.

The audience for this book is a doctor, nurse, pharmacist, biostatistician, or clinical informaticist who has watched the engineers at KHCC (or any other hospital) ship things, wondered how it was done, and wanted in.

You are in.

— I.S. Amman, 2026