Website and booking system for an upholstery cleaning business: AI builds the order from a photo right on the customer's phone, the price is agreed on WhatsApp and the calendar only offers slots that are actually free.
Project built for LIMPO.AI's Águas Claras location (Federal District, Brazil).
At a glance
Upholstery cleaning is usually sold over WhatsApp, with a lot of back-and-forth to send photos, explain what needs cleaning and find a time. The challenge was to shorten that path, with the order arriving ready in the chat and real availability right on the website, and to give the location a management system with no fixed costs, since it was just starting out.
We built a website where customers handle everything in a few steps: they take a photo and the AI builds the order, they adjust the items if needed and pick a free time slot. The order arrives ready on the owner's WhatsApp, and the price is agreed in the chat. Behind it, an admin panel with calendar, customers, pricing, promotions and marketing runs only on services with free plans.
What was delivered
Four steps of the website, as they are live.
Right at the top, visitors move the mouse (or a finger) over a stained cushion and watch the dirt disappear while a counter goes from 0% to 100% clean.
The customer sends a photo and the AI recognizes a sofa, armchair, chair or bed and builds the order. The analysis runs in the browser itself: the photo only leaves the device when the customer taps “Veja o preço” (“See the price”), and then it goes to the team along with the message.
The customer builds the order item by item and sees the on-site time and the drying time. The price comes from LIMPO.AI itself, on WhatsApp, with the order already written in the message.
Only free slots appear, Monday to Saturday, morning and afternoon. When the form is sent, the slot is checked again and the order arrives ready on the owner's WhatsApp.
Team admin panel
What appears on the website is edited in the admin panel and updates instantly, with no redeploy. Each person only sees the areas their access level allows.
The needThe team works with limited morning and afternoon slots and stops on days off and holidays.
What was builtA weekly calendar with free slots per period, blocking a whole day or one period with a reason, and service reminders with the WhatsApp message ready to send.
Fictitious dataThe needKnowing who came back, who disappeared and who referred someone, without a spreadsheet.
What was builtA CRM built from the bookings: filters for returning customers, customers with no order in 6 months and pet owners, plus tags, notes and total spent per customer.
Fictitious dataThe needPrices are agreed on WhatsApp, but the team needs a reference for each piece.
What was builtA table of items with price and time that never shows on the website: it is the team's reference and becomes the estimate in the calendar. A disabled item leaves the website without losing its price.
Fictitious dataThe needCampaigns and coupons change every month.
What was builtPercentage or fixed-amount coupons with dates, usage limit, minimum order and services, live on the website instantly. Referrals and the pet plan sit on the same screen.
Fictitious dataOn the phone
The needThe calendar is checked on the go, between one job and the next.
What was builtThe same calendar on the phone, day by day, with each period's bookings and the blocked times.
Fictitious dataOrders, revenue earned, average ticket, customer sources and the schedule for the next 7 days.
List or weekly view, with confirm, complete and cancel, the amount charged and blocked days.
A CRM built from bookings, with history, tags, referrals and ready-made WhatsApp messages.
The quote calculator table, installments, coupons, referrals and a pet plan, all editable.
Orders by channel and campaign, a UTM link builder, and GA4, Google Ads and Meta Pixel IDs.
Administrator and Agent levels, lockout after wrong passwords, email recovery and a change history.
Design in Figma
Technology
Next.js
Version 16. Website, admin panel and API routes in the same app, with a daily reminder job.
React
Version 19. Interactive components: the cushion, the quote calculator, the photo quote and the calendar.
TypeScript
Pricing, scheduling and permission rules typed and validated with zod.
Tailwind CSS
Tokens for the brand and the website's visual language.
MediaPipe
Object detection in the browser, in a Web Worker, with no AI server.
GSAP
Page animations and smooth scrolling with Lenis.
Prisma
Bookings, customers, settings and history in Neon's serverless Postgres.
Vercel
Deployed in São Paulo, with a daily Cron job for service follow-up reminders.
Engineering
The roughly 7.8 MB model is only downloaded when the section comes into view and runs in a Web Worker: the page never freezes and there is no AI API bill.
Each booking checks the slot in a serializable database transaction, so two customers never get the same time.
UTM, gclid and fbclid are kept for 30 days and travel with the booking. The admin panel groups orders by channel and campaign.
GA4, Google Ads and Meta Pixel only load after cookie consent, and a customer's data can be deleted from the admin panel.
Visual identity
Colors and fonts checked on the live site, not just in the code.
Palette · click to copy
Headings
Mancha sai.
Space Grotesk
Headings and highlight numbers.
Accents
Ácaro sai.
Instrument Serif
Italic words inside headings.
Body
Higienização no seu endereço
Manrope
Body text and forms.
Labels
PRÓXIMA VAGA: AMANHÃ
JetBrains Mono
Tags, counters and the quote receipt.
Next step
We build websites that put the order together and show availability on the spot, with an admin panel for your team to manage orders.
