Wellness Events App
Helping independent organizers publish wellness events straight from WhatsApp, and helping people find them by date, type and distance.
Side project
AI product
End to end
Native app
Built with Claude
THE PROJECT AT A GLANCE
3
flows designed end to end
B2B + B2C
one product for organizers and attendees
AI
builds the event from a WhatsApp message
PROJECT
ROLE
PLATFORM
BUILT WITH
Personal side project
Product Designer
Native apps &
WhatsApp
Claude
Context
Cities are full of small, independent creators offering wellness experiences: Breathwork circles, run clubs, dance classes, retreats and workshops. Most of them promote their events in WhatsApp groups, because sharing a post there is quick and free.
The idea came from my own experience. I’m in two of these groups, one for personal development and one for sport. New events arrive every week and get lost in the chat. As an attendee, they’re hard to track and find. As a creator, WhatsApp is still easier than a big events platform.
So I designed a product around what people already do, end to end, with Claude as my AI partner. AI is also at the core of the product.
Problem
Events live in WhatsApp chats, where they are easy to post and hard to find.
Organizers reach few people
Posts get buried within hours. Bookings happen one by one, by private message.
Attendees can't filter
No way to search by date, type, price or distance. Price and place are often missing.
Existing tools add friction
Some competitors send organizers to an external website. Others ask for a subscription first.
HOW MIGHT WE
How might we let organizers publish in one step from where they already work, and help attendees find and book the right event fast?
My role and how I used AI
Product Designer, end to end, on a personal project. Claude was my AI partner, from research to prototypes in code.
01
Framed the problem, principles, scope and business model
02
Analysed 18 real posts and benchmarked 11 apps
03
Designed the information architecture and three end to end flows
04
Built the design system and clickable prototypes with Claude, ready for code
How I worked with AI
Claude was my design partner, from research synthesis to prototypes in code. I set the direction and made every call.
What I owned
The problem, the principles, every decision and the review of each screen.
What Claude sped up
Synthesis, benchmark write ups, visual directions, copy and prototypes in code.
How I kept it honest
A brief, a decisions log and shared rules. No invented data: what isn't validated is labelled.
Discovery
Desk research
18 real event posts (55 screenshots) from a WhatsApp group. The poster often holds the date and place, so the AI must read images.
A benchmark of 11 competitors in Miro
ANATOMY OF A WHATSAPP EVENT POST
2
languages, often mixed
4
event types
5
price formats
5
ways to book
KEY INSIGHT
Organizers don't need another form. They already have the content. They need someone to turn their post into a page.
Approach
TODAY
Make a poster
Post in groups
Lost in the chat
Book by DM
Pay outside
WITH THE APP
Forward poster
AI builds it
Confirm
Set up payouts
Live
Publishing goes from scattered posts to one forward.
THE BRIEF: A WELLNESS EVENTS APP
My starting hypothesis
We believe that if organizers can publish from WhatsApp, where they already are, they will reach more people, and attendees will book more because they can find the right event more easily.
AI at the core
Every AI feature has one clear job, and a person stays in control.
AI features
1. Reads posters and text, asks only for what is missing
2. Rewrites health claims and sends risky events to a person
3. Suggests images and recommends events to attendees
Key decisions and trade offs
01 Publish by forwarding a post
Quick steps stay in the chat.
Trade off: Only 3 reply buttons of 20 characters.
02 The AI asks, it never guesses
Price and date are always confirmed.
Trade off: One or two more taps.
03 Payouts before going live
No money is ever held without a payout account.
Trade off: A heavier first step.
Design system
I explored three directions and chose "Clean and Modern": few colours, soft shadows, no borders. Built on shadcn/ui tokens, so it drops straight into code.
DIRECTION 1
DIRECTION 2
DIRECTION 3: CHOSEN
COLOUR TOKENS, NAMED LIKE SHADCN/UI
Mint
#B3EEC6 · Actions
Charcoal
#1E2624 · Text, prices
Deep teal
#1D7075 · Links
Muted
#5C6866 · Meta
Page
#F2F5F4 · Background
RULES I SET FOR EVERY SCREEN
1. No borders. Depth comes from soft shadows.
2. Selection is never colour alone.
3. 44 px touch targets.
4. Use Native components.
A KEY CHANGE
Flow 2 started as a web editor, opened from a link in WhatsApp.
Since the product is native, I turned it into a host mode in the app.
Iteration
AI’s first draft of the ‘create an event’ screen showed every single option at once. It felt crowded, so I redirected it to apply progressive disclosure: The two main actions first, the other options one tap away.
FIRST VERSION
Six ways to start on one screen.
FINAL VERSION
Two clear choices first: a post or an idea. Other ways sit behind a link.
WHY IT WORKS
Fewer choices, faster decisions.
Hick's law: Every extra option adds decision time. The quickest path comes first, without hiding the rest.
The three flows
FLOW 1:
Organizer
Upload an event from WhatsApp
The organizer forwards her usual post. AI reads it, asks only for what is missing, confirms the date, improves the text and image, then sets up payouts and goes live.
FLOW 2:
Attendee
Attendee app: Discover and book
A personal Explore, filters with counts, map and list on one screen, event pages with price and place up front, and checkout.
FLOW 3:
Host
Upload an event in the app
Organizers switch to hosting and start from a post, a poster, a text or a voice note. AI builds the draft, and an Ask AI bar changes anything.
Beyond the happy path
The happy path is one forward. The work was everything that can go wrong.
Duplicate "last call" posts
Reposts are caught, so the same class never shows twice.
Missing price or place
The AI asks one question at a time, within 3 WhatsApp buttons.
Few spots left or sold out
A "5 left" badge, read as "5 spots left" by screen readers.
How I will measure success
Time
From first message to a published event
Share
Of events published without human help
Tickets
Sold per event
BUSINESS MODEL
Free to list. Attendees pay a 5% service fee.
Organizers get the full ticket price. A hypothesis to test.
The data events
Defined before build, so the pilot answers real questions from day one. No data yet.
Event
Question it answers
EVENT_POST_FORWARDED
Will organizers forward events to a number?
AI_FIELDS_EXTRACTED
How much does the AI read on its own, from poster and text?
EVENT_PUBLISHED
How long from first message to a live event?
CHECKOUT_COMPLETED
Will attendees pay in the app, not to the organizer?
Risks to watch
RISK 1
Channel
Meta bans general purpose AI chatbots on WhatsApp. Ours does one task.
RISK 2
No user research yet
The biggest risk, and the first to test.
RISK 3
AI accuracy and trust
Messy posts in Spanish and Catalan.
RISK 4
Business
Payment outside the app and low ticket prices.
Next steps
STEP 1
Validate
A poll, and 5 organizers forwarding real events
STEP 2
Build
WhatsApp to a live event page, in Claude Code
STEP 3
Pay
Payouts and checkout with Stripe, in test mode
STEP 4
Native app
The native attendee app, from the prototype
Learnings
01
Start from the raw material
Real posts defined the AI's job better than any feature list.
02
Meet people where they already are
Organizers live in WhatsApp, so the product starts there.
03
AI needs product rules, not just prompts
"Ask, never guess" mattered as much as any screen.
04
AI speeds up design, not validation
Claude helped me move fast. Real users still decide what works.