Owning a real-estate WhatsApp backend

I owned the backend of a real-estate WhatsApp platform covering property questions, document ingestion and retrieval, structured lead summaries and site-visit booking.

For booking, I kept validation, storage and confirmation in a defined backend workflow. The model handled conversation, recognized booking intent and opened a WhatsApp Flow form to collect the visitor’s name, date and time.

Backend code validated the submission and returned defined feedback for invalid input. Valid submissions became booking records linked to the user’s context, followed by a predefined confirmation message.

I also introduced a delayed form-reminder workflow using Hatchet. A form-start event scheduled the reminder, and a form-submitted event cancelled the pending job. I tested that workflow locally and in the deployed environment.

The backend used TypeScript, Bun and Hono, with Postgres, Qdrant, LlamaParse and S3.