How I Built Two IB Exam Practice Tools Using AI

Let me be completely clear about my technical background before this post begins: I am not a programmer. I took one computer science class in 2003, I cannot write JavaScript from scratch, and until about eight months ago the words "CSS grid" meant nothing to me. I am a high school teacher who coaches basketball and runs a media program in Halifax.

That's context, not humility. Because what I built this spring should not have been possible for someone with my background — and the fact that it was possible is the actual point of this post.

My IB Digital Society students were preparing for their May exams. I wanted a self-assessment tool that worked the way I think about exam prep: structured steps, real rubric language, a place to plan before writing, and a way to annotate your own response to see what you've actually argued. Nothing like that existed. So I decided to build it.

I ended up building two.

What I Actually Built

The first tool — the one I started with — is called the IB DigSoc Exam Practice Studio. It's a single HTML file that runs entirely in the browser, no server needed. You select your paper type (Paper 1 or Paper 2) and the mark value of the question you're practicing. The app shows you the official IB rubric for that question type, a checklist of what a high-scoring response actually needs, an annotation system where you can stamp different parts of your written response with labels (DEF, EV, ARG+, ARG−, etc.), and a score estimator that updates as you tick off checklist items. When you're done, you can export the whole thing — your answer, your stamps, your checklist state — as a single file you can email to your teacher.

The second tool — which I built after the first one and is considerably more complex — is the DigSoc Response Studio. This one walks students through a seven-step process: understand the question, think like a high-scoring student, brainstorm before writing, write the full response, annotate it with inline text highlights, self-assess against the rubric, and then generate a tailored AI feedback prompt they can paste into Claude or ChatGPT. It covers every question type across Paper 1 and Paper 2, with specific rubric logic, checklists, and exemplar guidance for each. It also has a demo mode that loads a full worked example — a real IB-style question on Instagram and social class — with a complete model response, annotations already placed, and a checklist that shows exactly why the response hits the top band.

Both tools run entirely in the browser as single HTML files. No login, no server, no subscription, no data sent anywhere. A student can open one on a Chromebook in a school with no internet and it still works. That was non-negotiable for me.

How It Actually Gets Built: The Process

This is what most posts about "vibe coding" skip. They say "I just chatted with AI and it wrote the code!" as if the hard part is typing. The hard part is not typing. The hard part is knowing what to ask for and knowing when to push back.

Here's the actual process I used, which you could replicate for almost any kind of tool you want to build as an educator:

Start with the problem, not the feature list

My first prompt was not "build me an IB exam tool." My first prompt was something closer to: "My IB Digital Society students consistently get to an exam and don't know the difference between describing something and analysing it. They write, they self-assess that it's good, and they're wrong. What's the root problem here and what would a tool need to do to fix it?"

That conversation — not a coding session, just a thinking session — produced the architecture. The tool needed: step-based flow so students can't skip the thinking phase, specific rubric language from the actual markbands, a way to visually locate where in their writing the evidence actually is, and a score that feels real because it's driven by the rubric, not vibes. Everything after that was building toward those four things.

Build the content before you build the code

The rubric data — the official markband language for each of the eight question types, the non-negotiables, the examiner tips, the things weak students do versus what strong students do — I drafted all of that in a document first. I knew the subject. I could look at my IB resources and write those tables. Once that content existed, asking the AI to turn it into a working application was a much simpler conversation than asking it to invent both at once.

This is something I'd tell any teacher trying this: you have domain knowledge that the AI doesn't have. You know what your students get wrong. You know what the rubric is actually asking for. You know what a good response looks like versus a mediocre one. That expertise is the most valuable thing in the room. The AI is the implementation layer, not the expertise layer.

Describe behavior, not syntax

I cannot tell an AI to "add an event listener to the textarea that fires a debounced save to localStorage." I don't know those words. What I can say is: "When a student types in the brainstorm field, their work should save automatically — they should never lose their text if the tab closes." The AI translates that into code. My job is to describe the behavior with enough precision that the translation is accurate.

You don't need to know how to code. You need to know how to describe what you want with precision. Those are very different skills, and you probably already have the second one.

Precision matters a lot. "Make it look better" gets you nothing useful. "The score estimate should update immediately when a checkbox is ticked — not on page load, not on submit — instantly" gets you what you want.

8 Vibe Coding Tips That Actually Matter in June 2026

A lot of what's written about AI-assisted coding is either breathlessly optimistic or uselessly vague. Here's what I actually found useful across the process of building these two tools:

What Surprised Me About the Process

I expected the hardest part to be the code. It wasn't. The hardest part was writing the content — the rubric language, the examiner tips, the checklist items that are specific enough to be genuinely useful but general enough to apply to any question in the paper. That took domain knowledge, experience with where my students struggle, and careful thinking about what the IB is actually trying to assess at each mark level. The AI helped me refine the language, but the substance came from me.

The other thing that surprised me was how much I enjoyed the process of building. I've never thought of myself as someone who makes software. But there's something satisfying about describing a problem clearly and watching a working solution appear — especially when you can open it, hand it to a student the next morning, and see them actually use it.

The annotation feature in the Response Studio — where students highlight text in their own response and tag it with a stamp — took about twenty minutes of back-and-forth to get right. The export system that embeds everything into a self-contained HTML file took an afternoon. Neither of those things would have been possible for me without AI. Both of them are now genuinely useful in my classroom.

What It Means for Teachers

I want to be careful not to oversell this. These tools work for my students because I understand my students, my subject, and the exam they're preparing for. An AI could not have built them without me. But I also could not have built them without the AI.

What's changed is not that coding is easy now. What's changed is that the implementation gap — the distance between "I know what this tool should do" and "this tool exists" — has shrunk dramatically for people who have subject expertise but not technical expertise. That's most teachers. That's almost all of us.

The skills that matter now are: being able to describe what you want with precision, knowing your subject deeply enough to build something that's actually useful, and being willing to iterate — to use the tool with students, see where it fails, and make it better. Those are things teachers already know how to do. The AI handles the rest.

If you've been sitting on an idea for a tool, a resource, a practice simulator, a self-assessment rubric, or anything else that would help your students — now is actually the time to build it.

Chad Wadden

Teacher, department head, and basketball coach in Halifax, NS. Writing about education, technology, and performance every two weeks.