Back to Projects

Appcitas

107,750 appointments booked and counting. An AI WhatsApp agent for healthcare clinics across Latin America, running since 2015 on Node.js and AWS.

SaaS WhatsApp AI HealthTech Visit site

The 10-Year Journey

Appcitas started in 2015 when I was just getting started in software development. A simple idea: help Colombian clinics and service businesses manage appointments without expensive software.

The problem: Clinics, dental practices, beauty salons — all drowning in appointment chaos. Phone calls, WhatsApp messages, missed appointments, double bookings.

What Appcitas Actually Does

A B2B SaaS platform where patients book, cancel, and reschedule appointments directly through WhatsApp — powered by an agentic AI.

  • AI booking agent — Handles natural conversations in Spanish, books and cancels appointments using 7 custom tools
  • Full WhatsApp integration — Two-way messaging via Meta Cloud API, each business gets their own WhatsApp Business Account
  • Automated reminders — Daily scheduled Lambda sends WhatsApp template messages 24h before appointments
  • Multi-tenant — Each organization isolated by phone number, conversation history stored per-tenant in DynamoDB
  • Public booking pages — Whitelabel booking widgets accessible by organization slug
  • CX analytics — NPS, CSAT, NEV surveys collected post-appointment with keyword analysis

Technical Architecture

Fully serverless on AWS:

  • Lambda functions for API, webhooks, NLP processing, and CRM
  • SQS queues for async message processing pipeline
  • DynamoDB for conversation state (24h TTL, 20 message window)
  • PostgreSQL/Supabase for relational data via Sequelize
  • EventBridge Scheduler for CRM scheduled messages
  • Wompi payment gateway integration (Colombian payments)

The WhatsApp flow: Meta webhook → parse + tenant lookup → SQS → AI agent with tools → response queue → send via Graph API.

Evolution

  • v1 (2015-2017): PHP + MySQL. Classic LAMP stack.
  • v2 (2017-2019): Node.js + MongoDB. WhatsApp became the killer feature.
  • v3 (2019-2024): Node.js + PostgreSQL + AWS serverless. Rasa NLU for chatbot.
  • v4 (2024-Present): Agentic AI replaced Rasa. Full tool-calling agent.

What I Learned

  1. WhatsApp is king in LATAM — Email reminders had 20% open rates. WhatsApp has 95%+
  2. Side projects need patience — Took 5 years to get meaningful revenue
  3. Simple beats complex — The features that stuck were the simple ones
  4. AI actually works now — Agentic AI with tools handles edge cases Rasa never could

Built by juanships.com

I wrote about this: