Built it2025–2026 · AI on WhatsApp
Falaí: what I learned from a product I paused.
An AI assistant on WhatsApp for small businesses: scheduling, FAQs and customer service, 24/7.
- Role
- Defined and built
- Stack
- n8n, MongoDB, WhatsApp API, OpenAI
- Status
- Paused
The pivot
One number and one server per client.
- Clinic n8n · WhatsApp
- Salon n8n · WhatsApp
- Dentist n8n · WhatsApp
- Studio n8n · WhatsApp
Every new client meant a new instance to provision, pay for and keep alive.
The idea
Small businesses lose customers on WhatsApp when nobody answers. Falaí answers around the clock: it schedules, handles FAQs and serves customers, with an AI agent configured for each business.
What I built
A multi-tenant platform on n8n, MongoDB and the WhatsApp Business API. Each business keeps its own number and redirects customers with a link to Falaí’s central number; a code in that link tells the agent which business it’s speaking for.
The pivot
I started with one server instance and one WhatsApp number per client, fully automated by a provisioning script. It worked, but every new client added cost and moving parts. I moved to a single central number: simpler to run, cheaper to scale.
What validation looked like
I showed it to four dentists. All four liked it and wanted to try it.
Why I paused it
Time. A full-time job, a freelance AI project and finishing my degree, all at once. Something had to wait.
What I learned
Sell before you build. Four dentists wanted to test it, and I had no version ready to hand them. Five pilots would have taught me more than months of architecture.
Trust AI, but verify. I had an autonomous agent audit my server; its report flagged working parts as broken. I checked every item before acting on it, and that habit now runs through how I work with AI.
Write decisions down. Documenting architecture decisions kept me, and the AI, from undoing them session after session.