SchoonMaak Pro — Prince Festus

Overview

What SchoonMaak Pro does.

In 2 days, I built a fully automated booking system for Dutch cleaning businesses. It responds to WhatsApp and email enquiries with AI, collects booking details conversationally, saves them to a database, notifies the owner, and reminds the customer 24 hours before their appointment.

Enquiry handling, booking collection, notifications, reminders, and follow-up workflows are automated. Owners can monitor activity through the dashboard and take over conversations whenever manual intervention is required.

The problem

Why this market needed automation.

Small cleaning businesses in the Netherlands manage all bookings manually on WhatsApp. They reply to every message themselves, miss enquiries when they're busy cleaning, have no overview of their schedule, and lose customers to competitors who respond faster. There is no affordable done-for-you solution built specifically for this market.

Why I built it

Market gap. Shipped in 2 days.

I identified that many Dutch cleaning businesses rely heavily on WhatsApp for customer communication and booking management, making them a strong fit for automation. SchoonMaak Pro was built to validate that fit with a working, production-ready system.

System Architecture

How WhatsApp, email automation, AI, scheduling, and owner workflows connect.

Customer

Input Channels

WhatsApp Channel

Twilio Business API

Email Inbox

Gmail API · OAuth2

↓

Core Engine

Conversation Orchestrator

Node.js · Express.js · Webhooks · Scheduling

AI Decision Engine

Claude

claude-sonnet-4-6 · Conversational booking · Dutch + English

Booking Logic

Validation · Duplicate prevention · 5-min window check

Data & Integrations

Rate Limiter

express-rate-limit

→

Supabase

PostgreSQL · Bookings · Customers · Sessions

←

Human Takeover

Pause · Resume bot

Output Services

Owner Notification

Instant WhatsApp alert

↓

Owner Dashboard

Bookings · Revenue · DM · Takeover

Reminder Scheduler

node-cron · 24h before

↓

Customer Reminder

WhatsApp · Appointment confirmation

Monitored 24/7 · UptimeRobot · Render.com

How it works

Three systems. One automated pipeline.

WhatsApp Flow

01Customer messages the business WhatsApp number

02AI responds instantly in Dutch or English

03Collects service type, home size, date, time, and address conversationally

04Confirms booking with full details

05Owner receives instant WhatsApp notification

06Customer receives a reminder 24 hours before their appointment

Email Flow

01Customer emails the business

02System reads it via Gmail API

03Claude AI replies intelligently — handling both enquiries and booking requests

04Booking saved to database → owner notified on WhatsApp

Owner Dashboard

01Password-protected URL — no login system required

02Total bookings, today's bookings, monthly revenue overview

03Full bookings table with customer details

04Take Over button pauses the bot — owner replies manually. Resume Bot hands control back to AI

05WhatsApp DM button — message any customer directly from the dashboard

06Monthly revenue breakdown. Fully mobile responsive

Demo

See it in action.

What you'll see in the demo

✓ WhatsApp conversation with the AI assistant

✓ Booking details collected conversationally

✓ Booking saved to Supabase automatically

✓ Owner notification workflow

✓ Dashboard updates in real time

✓ Human takeover and resume workflow

Outcomes

Where it stands.

2

Days — built and deployed

24/7

Running on Render

3

Channels — WhatsApp, Email, Owner Dashboard

End-to-end booking workflows were successfully tested, including enquiry handling, booking creation, owner notifications, and automated reminders. The platform is deployed and ready for customer validation. Customer acquisition is ongoing.

Validation status

Where this project stands.

The platform has been deployed and tested end-to-end. Customer validation and outreach are ongoing.

Engineering challenges

What was actually hard to build.

01

Twilio sandbox delivery failures

Responses were being generated correctly but never delivered. The root cause was Twilio sandbox international restrictions and daily limits affecting Nigerian numbers during development — required moving to a production number and restructuring the delivery flow.

02

Server crash on new customers

A .single() Supabase call was throwing an unhandled error when no booking existed for a new phone number — silently crashing the server on every new conversation. Fixed by switching to .maybeSingle() with explicit null handling.

03

Duplicate booking prevention

Claude would sometimes generate the BOOKING confirmation tag multiple times in edge cases. Field-matching alone wasn't reliable — solved with a 5-minute time window check that prevents duplicate bookings from the same number within the same session.

04

Railway SIGTERM handler bug

Adding a SIGTERM handler for graceful shutdown caused the server to shut down on every incoming request in Railway's environment. Required removing the handler entirely and relying on Railway's native process management.

05

Dutch date accuracy

Claude was calculating booking dates incorrectly because it had no awareness of today's actual date. Fixed by injecting the real-time Netherlands timezone date into every system prompt call — ensuring all date calculations are grounded in the correct local time.

06

Conversation memory leak

Storing unlimited conversations in memory caused a potential memory leak on long-running servers. Fixed with hourly cleanup that removes all conversations older than 24 hours — keeping memory usage stable across uptime.

Built with

The stack behind it.

AI

Claude API (claude-sonnet-4-6)

Backend & Runtime

Node.js Express.js node-cron express-rate-limit

Integrations

Twilio WhatsApp Business API Gmail API (googleapis) Nodemailer

Database & Deployment

Supabase (PostgreSQL) Render.com UptimeRobot GitHub

What this tells you

What I bring to a team.

SchoonMaak Pro demonstrates the design and deployment of a production-ready business automation platform integrating conversational AI, messaging systems, scheduling, customer notifications, human takeover workflows, and operational dashboards.

The project was built and deployed in two days, with a focus on reliability, usability, and real-world business workflows rather than a prototype-only implementation.