WhatsApp Messages Turned Into Property Listings

The situation

In Mumbai real estate, inventory moves through WhatsApp. Brokers post flats for rent and sale, and buyer requirements, into groups all day, in their own shorthand: “2bhk khar w 4th flr 85k semi furn”.

The AV Properties team read these messages and retyped the useful ones into the listing website. It was slow, listings were missed, and the same property was often entered twice.

What we built

A backend service that connects to the team’s WhatsApp accounts and does the reading for them.

• Each team member links a WhatsApp account by scanning a QR code and chooses which groups and chats to monitor.
• Every incoming message is sent to an AI model that returns the listing as structured data: listing type, locality, configuration, price, area, building, furnishing and floor.
• The model works only from what the message says. A field that is not mentioned stays empty rather than being guessed.
• Prices written as “1.5 cr”, “50 L”, “85k” or a per-square-foot rate are converted to a plain number, with a sanity check for the way brokers use “k”.
• For each listing the model writes a title, a full description and four to six FAQs for search.
• Test messages, spam and anything that is not a property are skipped, and a message already processed is not processed again.

Keeping broker details private

Broker messages carry names and phone numbers. The agency wanted enquiries to come to it, so the public title, description and FAQs never include them. Contact numbers are stored in a separate field for the team only.

The result

Finished listings are placed on a queue for the agency’s listing website, and the team can export everything to Excel. A dashboard shows each WhatsApp connection live, and token usage is logged per message so AI cost stays visible.