DAO
Back

automation

Kosan Pagar Putih — Boarding House Automation

End-to-end WhatsApp automation for a boarding house: lead capture, rent billing, reminders, cash tracking, and an owner dashboard, all synced to Google Sheets.

Role
Architecture, workflow design, deployment and operations by me; code built with AI assistance
Period
Oct 2026 – present, in daily use
Stack
n8nWAHA (WhatsApp HTTP API)GeminiGoogle Sheets APIDockerLinux VPSFastAPIReactTailscale

Results

  • Replaced manual WhatsApp follow-ups with an automated lead pipeline
  • Monthly rent bills and reminders are created and sent automatically
  • Single source of truth in Google Sheets, read/write via a custom dashboard
  • Weekly owner summary generated automatically

Problem

A boarding house with multiple rooms was managed entirely by hand: inquiries answered one by one on WhatsApp, rent reminders sent manually, payments tracked in a spreadsheet, and the owner had no quick overview of occupancy or arrears.

What I built

Running in daily use at a real boarding house since October 2026 (the demos below use sample data). A self-hosted automation stack on a Linux VPS:

  • WAHA bridges WhatsApp to an HTTP webhook.
  • n8n orchestrates workflows: lead qualification, billing, reminders, cash tracking, weekly summaries.
  • Gemini classifies and replies to tenant messages in natural language.
  • Google Sheets acts as the database (leads, tenants, billing, cash flow, config).
  • KosPanel, a web dashboard, reads from and writes back to Sheets so the owner can manage everything from a browser.
  • Owner commands over WhatsApp: record payments, cash in and out, closings, vacating a room, and request summaries without opening the dashboard.

How the data flows

Data flows: WhatsApp → WAHA → n8n → Gemini (for understanding) and → Google Sheets (for persistence). KosPanel reads/writes the same Sheets.

See the architecture diagram (sample data, sanitized).

Dashboard demo

KosPanel demo: occupancy, room status, billing and cash, read live from Google Sheets.

WhatsApp chatbot demo

  1. Availability from live data. Asked about a free room, the bot reads the sheet: every room is taken, one frees up on a set date, and it says exactly that.
  2. Consistent answers to common questions. Price, what is included and location come from fixed templates, so the wording never drifts.
  3. It knows when to hand off. For an ambiguous question the bot asks what the person means and sends the owner a short summary to follow up on.
  4. No pressure. When the prospect says they will survey later, the bot replies politely and leaves the lead open for follow-up.

Recorded on a phone from the prospect's side and edited down: waiting time between messages is cut. The owner alert appears in the same chat only because this demo uses one phone for both roles.

Architecture

Challenges and decisions

  • Reliability. The bot first ran on a laptop and went down whenever it was offline. I moved it to a Linux VPS with automatic restart, scheduled backups, and a tested reboot.
  • One message, one alert. A single chat message could trigger two owner notifications. I moved the decision to after the reply and reused one tag from the same AI answer, so there are no extra AI calls.
  • Safe edits from two directions. The dashboard and WhatsApp can both change the same sheet. Every web edit carries the old values and is rejected if the row changed meanwhile, and each edit is logged.
  • A human stays in control. When the owner replies manually, the bot pauses on that chat for 6 hours, and payment proofs wait for the owner's confirmation before they are recorded.