Designers7 min

The design thinking process, explained with a real project

The design thinking process explained with a real project: how empathise, define, ideate, prototype and test played out on Pukaar.ai, a baby care app.

Saurabh YadavFounder, Design Studio by Saurabh
Designers
Share

Search for "design thinking process" and you'll find the same five hexagons everywhere: empathise, define, ideate, prototype, test. Neat diagram, sticky notes, a smiling team around a whiteboard.

The stages are real and useful. But the diagrams make it look tidy and linear, and real projects are neither. You loop back, skip ahead, and sometimes a single user quote rewrites a week of work.

So instead of another abstract explainer, I'll walk through how the five stages actually played out on Pukaar.ai, a doctor-led co-parent for a baby's first year that I worked on as UX researcher and interaction designer. We went from zero to live in 15 weeks, with a team of a CEO, CPO, CTO, Head of AI and 10+ paediatricians.

A quick refresher on the five stages

The model most people learn (and the one I studied in my PG Diploma at AND Academy) has five stages:

  1. Empathise: understand the people you're designing for.
  2. Define: turn what you learned into a clear problem statement.
  3. Ideate: generate many possible solutions.
  4. Prototype: make cheap, testable versions of the best ideas.
  5. Test: put them in front of real users, learn and refine.

On Pukaar, my own process was labelled a bit differently: Research and Discovery, Define and Strategy, Wireframing and UX Design, UI and Visual Identity, then Insights. Same spirit, just names that matched how the team worked. Don't get hung up on the labels. Get hung up on the habits.

Here's the whole project on one page before we go stage by stage:

Stage The question it answers What it looked like on Pukaar.ai
Empathise Who are we designing for, and what do they feel? 70+ interviews and a WhatsApp community of real 2 a.m. questions
Define What exactly are we solving? Aarti, "The Anxious First-Timer", at the fourth wake-up of the night
Ideate What are all the ways we could solve it? Four ways to help, and the record-first flow
Prototype What's the cheapest version we can try? Wireframes to the Cry Report, 5 areas and 29 screens
Test Did it work, and what's next? Live on both stores, reviews feeding the next round

Empathise: go where the real problems are

The problem sounded simple: babies cry to express everything, from hunger to pain to tiredness, and parents struggle to understand what a cry means. That gap leads to anxiety, confusion and sometimes delays in care.

But "parents are anxious" isn't something you can design for. We needed the texture.

What we did:

  • 70+ user interviews with parents.
  • A WhatsApp community of 400+ parents, with 50+ queries a day, real 2 a.m. questions and polls.
  • Competitor research: no direct competitor in India and one globally (NannyAI), with indirect ones like Mylo, Healofy and BabyChakra.
  • Desk research on postpartum anxiety and infant health in India.

The WhatsApp channel taught me the most. Interviews tell you what people remember. A 2 a.m. message tells you what they actually feel in the moment: short sentences, no punctuation, pure worry.

If you take one thing from this stage: find a place where your users talk honestly when you're not in the room. I go deeper on this in research first, pixels second.

Define: one persona, one moment, one problem

All that research needs to collapse into something sharp, or it's just a pile of notes. We used affinity mapping to group what we heard, then built a persona.

Aarti Sharma, "The Anxious First-Timer". 29, lives in India, software engineer, first baby (three months old), Android and WhatsApp first. It's 2 a.m., the fourth wake-up of the night.

The persona came with three needs:

  • A plain sentence first.
  • Reasoning she can check.
  • A visible way out: soothe, log or consult.

That became our problem framing: how might we give a frightened, exhausted parent a clear, trustworthy answer about her baby's cry, fast, with one hand?

Notice how specific that is. "Improve parenting experience" would have given us nothing to decide with. A good problem statement should make some ideas obviously wrong.

Watch out

If your problem statement could sit on any app's pitch deck, it's too vague. Add a person, a moment and a constraint, like Aarti, 2 a.m. and one hand.

Ideate: lots of options, then ruthless cuts

With the problem defined, we explored widely. The vision was to start with a cry analyser and smart health screening and grow into a full AI platform supporting parents day and night. That became four ways the app helps: the Cry Analyser, AI Assessment, Experts Support and Educating Parents.

The more interesting ideation happened inside flows. The key question was: what's the fastest path from a crying baby to a useful answer?

One early idea was to ask a short questionnaire before recording the cry, so the AI had context. Logical on paper. But held against Aarti at 2 a.m., it fell apart. Asking four questions before recording was homework while the baby screamed.

So we flipped it: record first, then answer what you can see. That single change came straight from the persona, and it's a good example of why the Define stage matters. Without Aarti, the questionnaire-first version would probably have shipped.

My tips for ideation:

  • Generate options on paper or in FigJam before opening Figma.
  • Judge every idea against the persona's worst moment, not an average day.
  • Kill your favourite if it fails that test.

Prototype: from wireframe to the Cry Report

We moved into wireframes and then high-fidelity UI with a design system built on the Satoshi typeface and a purple outline icon library. The information architecture covered 5 areas and 29 screens.

The Cry Report screen shows how a prototype evolves. Between wireframe and final UI:

  • Reassurance moved above the actions, so the first thing a parent sees calms them.
  • The category label ("Needs Soothing") was demoted below a plain-language explanation.
  • Reasoning got its own block, "Why I think that", reading back the parent's own sleep and feed history.
  • Escalation stayed always visible, so a consult with a paediatrician was never more than a tap away.

The flow target was cry to answer in under 90 seconds. Every screen in the 2 a.m. path was designed one-handed, with oversized tap targets, no blocking modals and primary actions in the bottom third. I wrote about those choices in designing for the worst moment.

Note

Prototype and test weren't two separate phases here. Each version of the Cry Report was shown, questioned and changed before the next one, which is how the loop is meant to work.

Test: the loop doesn't end at launch

Testing is the stage people treat as a final exam. It's really the start of the next loop.

Pukaar went live on Google Play in March 2026 and on the App Store in April 2026. It has 50K+ downloads on Google Play and 4.3 stars from 170 ratings. Reviews from parents praise the cry analyser and QARA, the AI companion, with lines like "less anxiety since I started using" and "an ask-away anytime doctor".

That's encouraging, but launch data also shows what to improve. My list of what I'd design next:

  • Show the top two cry reasons with confidence, instead of one answer.
  • Start listening right away with a live waveform, instead of a 10-second wait.
  • Let parents try a cry analysis as a guest before signing up.

Tip

Read your app store reviews like interview notes. Group them the same way you grouped research, and you'll often find your next problem statement waiting there.

Each of those goes right back to Empathise and Define. That's the loop working as intended. If you're running tests with a small budget, my guide to usability testing on a budget shows how.

Work with us

Need a hand? Our product design agency works with startups and product teams from research to developer handoff.

Key takeaways

  • The five stages are habits, not a strict sequence. Expect to loop back.
  • Empathise where users talk honestly, like a WhatsApp community, not just in formal interviews.
  • A sharp persona and problem statement should make some ideas obviously wrong.
  • Judge ideas against the user's worst moment; that's how Pukaar moved the questionnaire behind the recording.
  • Launch is a test too. Real reviews and data feed the next round.

Use the process, don't perform it

Design thinking isn't about sticky notes or workshops. It's about staying close to real people, being precise about the problem and changing your mind when the evidence says so.

You can see every stage in detail, with screens, in the Pukaar.ai case study. If you'd like to ask questions about the process, try Sage, my AI assistant, or tell me about your project.

Saurabh Yadav's Memoji

Written by

Saurabh Yadav

Founder of Design Studio by Saurabh, a UI/UX design studio in India. We design apps and websites that feel effortless, from research to shipped pixels. Most recently: Pukaar.ai, as seen on Shark Tank India, and the Demerara Bank app.

Keep reading

All articles →