Building With Claude: From Simple Prompts to a Working Prototype

Continuing from my previous series on Claude and experiments - Click here for post #1

In my experience, Claude has been remarkably versatile in understanding natural language. I simply tell it what i want to do, and it gets the instructions quite clearly. This is obviously the easiest and path of least resistance as a newbie. But if you’re serious about learning AI, you should also be serious about learning the techniques that go with it.

Definitely not simple!

When I first started using Claude, I kept things simple - tasks like asking it for stock prices or helping me summarize tasks. Basic stuff most product managers already do with their favorite AI tools. But to truly harness Claude’s power, you have to go beyond these surface-level productivity hacks. You have to learn by doing.

Learning by Building

Here’s a free course I found useful:

AI Fundamentals for PMs.

The beauty of this course is that it lets you practise as you learn with real world examples and I found that this is a perfect way to understand how AI reasoning fits into real-world product workflows.

I downloaded the course and decided to follow along but whats the fun in just re- doing what the tutor said would most certainly work! So i decided to go with a twist. I wanted to apply these learnings to a personal project I’ve been sitting on for a while and something that i found would be useful for me as a mom of a teenager.

For the purpose of this learning i will not provide instructions on how your user personas should be set up or how interviews should be conducted etc. I assume you would be familiar with these general PM fundamentals.

I decided to use Claude to generate an app called Gyan Guru.

From Idea to UX Prototype

Think of Gyan Guru as an “Uber for tutors.” It connects three groups:

1. Students or caregivers looking for personalized classes

2. Tutors who want to teach online (1-1 or group formats)

3. Educators who already have video lessons and want to reach more learners

   
   

I’ll share user personas and research notes in a separate post, but for now, let’s stay focused on Claude. I already had a low-fidelity design in my notes, so my first step was to ask Claude to analyze it without giving any background on the app.

Low Fidelity Hand drawn screens

I wanted to see how it interpreted the design on its own. The goal was simple: test whether Claude could provide actionable UX feedback purely from visual input.

Note: These are simple low- Fid sketches i already had ready but did not have high- Fid prototypes available to kick start. I also asked Claude to go back and offer me suggestions on my design, did i add my carousel right, how is the placement of the confirmation panel etc.

Tell me what I did wrong!

Once Claude offered recommendations, I pushed it further iterating toward a high-fidelity prototype. This is where your product instincts really matter. AI can suggest improvements, but only you can decide which ones align with your vision. In my case, I skipped the idea of gamified badges (it felt dated) and focused instead on information hierarchy and a mobile-first design. A few iterations later — and about 45 minutes of back-and-forth — I had a fully refined UX layout.

Lovely to see the design come to life!

This is not by any means the final product, but only intended to tell you how quickly you can go from thought to validate.

Lessons Learned

Even with Claude, the fundamentals don’t change. You still need to do your competitive research, define your user journeys, and validate your assumptions. What Claude did was make it faster to visualize what I was thinking and turning vague concepts into tangible screens almost instantly.

It wont replace you right away , but if you dont learn to use these tools the way they were intended to be used, you might as well be replaced sooner by someone who knows how to.

I will post about my experiments in writing a pdf parser using Claude next, stay tuned!