Answers in the customer’s language
Georgian, Russian or English, on WhatsApp and other messaging apps. It keeps the conversation in the language the customer writes in.
Our work · AI agent · Real estate
It answers buyers and owners on WhatsApp and other messaging apps and on the phone, in Georgian, Russian and English. Behind it sits a database of the market that it reads, cleans and scores on its own. We built it and we run it.
Machine work, read from its own logs
Counts of machine work, read from each system's own database in Oct 2026. They include our own testing and they are not sales results.
One agent, one set of rules, whichever channel the customer uses.
Georgian, Russian or English, on WhatsApp and other messaging apps. It keeps the conversation in the language the customer writes in.
A voice agent answers calls on the agency’s number and works from the same listings and the same rules as the chat agent.
Looks flats up by what the customer asks for and sends each match as a listing card in the chat.
Notes what the client is looking for, asks them to confirm it, and uses it the next time they write.
Proposes time slots, checks them with the owner of the flat, then confirms the time with the client.
Collects the details of an owner’s flat in the chat and records them as a listing intake for the agency.
Hands the chat to a broker on request, or asks a broker when it does not know an answer, and carries the reply back.
The agent is only as good as the listings behind it, so this part runs without anyone.
Full and incremental reads of both portals through the day. Most of the 275,735 adverts it has seen are live, and each one is stored once.
Photo fingerprints on 263,538 of those adverts, plus the listing details, tell it that two adverts show the same flat, so it is counted and offered once.
Every active flat is compared with comparable flats in its area, 75,364 scored across 1,543 comparison groups, so a price can be called high or low.
A nightly job scores new flats against open buyer requests, keeps the ones above a set threshold and sends the customer a notice.
Both are built into the system. We list what they do and show no numbers for them.
Rent, lease, deposit, loan or mortgage, price change, term change and termination, filled in from the deal’s details.
Queues a listing for the property portals and for a Facebook Page and Instagram through their official APIs.
An agent that talks to customers needs a way to find its own mistakes.
158 scripted test cases check the agent’s rules and tools, and 37 simulated customers talk to it while a judge scores each transcript.
After calls end, a review measures speed and outcome, judges the transcript and lists the top problems with a quoted line and a proposed fix. It proposes only. Nothing changes on its own.
The other system we run is our own sales engine: it finds, audits and contacts local businesses.
How the sales engine worksFree 30-minute call
Bring the tools you use and the steps your team repeats. You leave the call knowing which system is worth building first, roughly what it costs, and whether we are the right people to build it.
Pick a time on the calendar. It is a 30-minute call, and the invite lands on your calendar when you book. Prefer email? [email protected]
We reply by email within 24 hours.
Checking the calendar…