An AI app prototype in a few days: how it works in practice

Why a prototype when there’s already a presentation

A presentation shows an idea; a prototype lets people test it. At Adamed the starting point was one diagram and a business direction. That wasn’t enough to decide on a tool for hundreds of consultants.

The decision makers wanted to see how an AI assistant answers questions from their everyday work. So I build a version people can use: click, ask a question and see the answer.

How I do it, step by step

1. I agree on the decision the prototype should help with. At Adamed, it was whether and how to build a tool for consultants.
2. I shape the product around one person’s working day. For MAIA, that’s a consultant’s day: doctor profiles, visit planning, product knowledge and training.
3. I collect the questions that person asks at work, and test the assistant on them.
4. I build a working version with AI tools, on demo data. I don’t draw separate mockups.
5. User roles and data security are in from the start, because the prototype has to open a conversation about a rollout.
6. I hand the prototype to the people who decide and collect their questions.

What I need from the company

• A description of the decision the prototype should help with.
• Someone who knows the process and answers questions during the build.
• Example questions and situations from the future users’ work.
• A go-ahead for demo data that resembles the real thing.

What the company has after a few days

A working prototype that anyone can open in a browser and try for themselves. At Adamed, the people reviewing the project asked the assistant about doctor profiles. On that basis, the company is discussing a rollout for about 700 consultants working at the same time.

A prototype isn’t a finished product. MAIA stays a prototype until the rollout decision. Its job is to shorten the way to that decision.