In my previous essay, I described how I began building clinical tools without writing code.
This follow-up is the practical answer to the question that comes next: what, exactly, do you type, and what do you do with the result?
“Describe what you want” is not enough instruction. Neither is “check that it works.” A physician needs a sequence: choose a suitable problem, establish the source, specify the behavior, obtain a working file, challenge its answers, correct its failures, and decide whether it is ready for anything beyond a demonstration.
That is the recipe below. You will not need to learn programming syntax. You will need to make decisions that an AI assistant should not make silently on your behalf.
The first decision is to keep the project small. Your first build should do one understandable thing, using a rule you can check independently. It should not diagnose a disease, choose a drug dose, predict a rare complication, or decide when a woman should deliver. Start with a calculation whose inputs, arithmetic and expected answers are explicit.
For this tutorial, we will build an adult BMI arithmetic demonstrator. It is deliberately modest. The aim is to learn the complete process on a formula that can be checked with a handheld calculator, rather than conceal a difficult clinical model behind an attractive screen. The demonstrator will show a number and the calculation. It will not classify a patient, interpret pregnancy weight, or recommend treatment.
The important distinction is that you are building a prototype. A page that opens, calculates and looks professional has cleared only the first hurdle. Agreement with known examples is an implementation check. A predictive model’s clinical validity, suitability for a population, and performance after deployment are separate questions.
There is also a correction worth making explicit: using AI does not transfer every responsibility to the clinician or remove the responsibilities of software developers and deploying institutions. Your clinical specification and review matter. So do competent software review, appropriate governance and clear ownership when a tool is used in care.
Use invented inputs throughout this exercise. Do not upload patient records or enter identifiable patient information. Keep the prototype out of clinical decision-making while you learn the process. These boundaries belong in the free introduction because they apply before anyone begins.
What paid subscribers receive below
The full tutorial includes the exact specification prompt, the build prompt, the file-opening instructions, an independently calculated answer set, a precise repair prompt, and a release checklist. It also explains what to do when the AI gives you code instead of a file, when a revision breaks something that worked, and when a risk model is too incomplete to implement responsibly.
If the first essay made you think, “I could build something useful,” this is the next step.
Become a paid subscriber to work through the complete recipe and support ObGyn Intelligence’s independent, evidence-focused work.
