Smart Consultant
An AI sales consultant running on your website and your WhatsApp line. It draws on a single source of information and brings the buyer to the point of a call back. Live today.
Estate Pilot is a sales infrastructure built for residential developers. At its core sits an AI consultant that reads your price list and unit data directly from the database. It does not quote prices from memory, does not invent square metres and does not negotiate. It greets the buyer, qualifies them, gives accurate information, opens a record and leaves the closing to a human.
An AI sales consultant running on your website and your WhatsApp line. It draws on a single source of information and brings the buyer to the point of a call back. Live today.
Every agent gets a permanent serial number and their own link. Who brought the buyer is recorded on the first click and cannot be reopened for debate later. Live today.
A matching layer where developers enter their projects and commission rates and agents see them in one place. Not built yet and not part of the current package.
Two separate leaks that need two separate answers.
A buyer's interest does not follow office hours. A question that arrives at midnight goes unanswered until morning, and by then the buyer has looked at another project.
Answering does not settle it either. Each sales representative quotes from memory: sometimes from an old list, sometimes rounded. If the price is set in euros, the lira equivalent drifts during the day and two representatives quote two different figures for the same unit.
On the agency side the problem is older. Who brought the buyer is often not recorded anywhere. That uncertainty produces commission disputes and keeps agents at arm's length from the project.
Estate Pilot addresses these two points separately. Information comes from one source, and who brought the buyer is recorded on the first click.
The consultant works in two places: the chat window on your website and your WhatsApp line.
It greets the buyer, asks what they are looking for and qualifies them: number of rooms, budget range, owner occupation or investment, and timing. It asks at most two questions per message and keeps sentences short.
When asked about price it returns the recorded list price. The list is fixed in euros; the lira equivalent is calculated with the live rate at the moment of each question, and the buyer is told the lira figure moves with the rate. If the rate cannot be fetched, only the euro price is given and no lira figure is invented.
Turkish is the default. Buyers who write in English or Russian are answered in their own language.
Once interest is clear it asks for a name and a phone number. Immediately before asking for the number it states the purpose of the record in one plain sentence, then opens the record and closes the conversation.
The consultant does not free-write; it works through defined tools. When a tool cannot answer, the conversation is not cut short: the consultant says the information is not available and hands the topic to a person.
Returns the official list price by unit type together with the lira equivalent at the day's live rate.
Queries the inventory by rooms, price range, floor and sales status.
Returns every field of a single unit by its number.
Name, location, concept and target handover date.
Returns only renders registered to the project; it never invents a link.
Name, phone, room preference, budget, purpose, timing and conversation notes are written into a single record.
On topics it cannot answer it moves the record to the human handover stage and writes down the reason.
The risk in sales software is less what it says than what it can invent. These limits are fixed not only as written rules but through executable behaviour tests.
It never gives price, area, floor or status from memory. The single source for these fields is the database.
It gives no discount, launch or campaign price, and never even implies one exists. Negotiation is handed to a person.
It makes no commitments on financing, interest, title deeds or legal process.
It shares the handover date only as a recorded target, never as a guarantee.
It does not comment on or disparage competing projects.
It does not hide being an AI when asked.
It does not invent what it does not have. It hands the topic over and tells the buyer plainly.
It never sends unsolicited promotional messages. It only replies to conversations the buyer starts.
Every agent has a permanent serial number of their own and a link carrying that number. When a buyer arrives through it, an attribution record opens and a durable trace is left on the device. The trace is valid for ninety days and cannot be read by the browser.
If a valid trace exists on the device, a link arriving later cannot overwrite it. The serial number does not change even as the link is passed from hand to hand.
Each device receives its own sub-trace, so how widely the link was forwarded stays visible.
A forged trace carrying a serial number that does not exist is never bound to a record. If a trace is lost for any reason, the system repairs it with the same serial number.
The system does not touch commission. It makes no payment, sets no rate and does not stand in the middle. It only documents who brought the buyer. The commission relationship stays entirely between the developer and the agent.
Estate Pilot is multi-tenant. More than one company's project lives in the same database. The critical question is this: under what circumstances can one company's data become visible to another.
The answer: it cannot. The boundary is drawn not in application code but in the database itself. PostgreSQL row level security is enabled on every table. Application queries run under a restricted role inside a single transaction. If the company identity is not set in that transaction, no row is visible at all.
The default is closed, not open. Even if the code has a bug, one company cannot pull another's rows; the database refuses.
The system creates a temporary test company, verifies in both directions that neither company can see the other's units, then deletes the test data completely. It runs with a single command, as many times as you like. We run it in front of you during the evaluation call. Not a slide, live output on screen.
Decisions that answer the security and data protection questions before they are asked.
The system runs entirely on Amazon Web Services in the Frankfurt region. Data stays within European Union borders.
The AI model is called through Amazon Bedrock. Both the model calls and the customer data stay inside the same AWS account. No key is handed to an outside AI service.
Keys and tokens are never written into code; they live in AWS Secrets Manager. The database disk is encrypted and backed up nightly.
A message arrives from Meta, its signature is verified, it is queued and a separate worker replies. The same message is never processed twice, so the buyer never gets the same answer twice.
All dates are kept in UTC and all monetary fields at cent precision. Rounding error does not accumulate.
Why each resource exists is written in the repository, and the cost table is updated with every new resource. A cost alarm is in place from day one.
A measurable answer to the question of trusting an AI.
The consultant's behaviour is put through a fifteen scenario examination. The exam uses the same setup that runs in production and makes real model calls. What is tested and what talks to buyers are the same thing.
Every price and every area figure in the answers is compared against the database exactly. A single number with no counterpart in the records counts as a failure.
The scenarios include: a buyer asking for a ten percent discount, a buyer pushing for a final price, a maintenance fee question, a buyer demanding financing and interest guarantees, a buyer raising a competing project for comparison, buyers writing in English and Russian, an image request and two end to end lead flows.
The lead flow additionally checks that the personal data sentence appears before the phone number is requested. If a single scenario fails, the release does not ship.
The commercial relationship is set out plainly.
Estate Pilot is used on a monthly software rental. Our revenue is the rental.
We take no share of sales. We take no share of commission. We do not enter the payment between you and the agent. The system records who brought the buyer; the rest is your commercial decision.
This choice is deliberate. A fee tied to sales volume makes the software a partner in the sale and creates a conflict of interest with both you and the agent. A fixed rental keeps the system neutral. The condition for trusting the attribution record is that the party keeping it takes no share of the outcome.
Infrastructure costs are transparent. What each AWS resource costs per month is laid out in a table, and what happens as usage grows is discussed in advance.
Three things are needed to start: your current price list, project information and the images you want to share.
The price list is loaded into the system and becomes the single source. The consultant speaks only from it. When the list is updated the behaviour changes immediately; nobody needs retraining.
Serial numbered links are generated for your agents and a chat window is added to your website. If the WhatsApp line is to be connected, Meta verification starts in parallel; that step can take a few days, so it begins on day one.
Before launch the behaviour exam is run, the isolation proof is executed and the legal texts are settled. The system is not opened to real buyers until all three are complete.
Said upfront, so nobody wastes time.
If your sales approach changes the price depending on the buyer, this does not fit. The consultant knows one list price and does not negotiate.
If you expect the AI to close the sale on its own, it does not fit. The consultant brings the buyer to the point of a call back; a human picks up the phone and a human signs the contract.
If inventory and pricing are not kept in order, the system will not fix that, only make it visible. Whatever the data says, the consultant says.
If you sell single units or resale property, the scale does not fit. The system was designed for selling many units within one project.
It cannot. The single source for pricing is the recorded list and the consultant does no arithmetic of its own. Producing a figure outside the list counts as a failure in the behaviour exam run before release, and that version does not ship. When the exchange rate cannot be fetched it gives no lira figure, only the euro price.
The consultant does not negotiate and never implies that a discount, launch or campaign price exists. It hands the topic to a person, writes the reason into the record and tells the buyer a consultant will clarify. Your sales team sees what was discussed the next morning.
It cannot. The separation is enforced not in application code but in the database itself, through PostgreSQL row level security. Without the company identity set, no row is visible. There is an executable proof of this and we run it in front of you during the evaluation call.
On Amazon Web Services in the Frankfurt region, that is, within European Union borders. The AI model is also called through Amazon Bedrock inside the same account; no data or key is handed to an outside AI service.
No. The system never touches commission: it makes no payment, sets no rate and does not stand in the middle. It only records who brought the buyer. The commission relationship is entirely between you and the agent. Our revenue is the monthly software rental alone.
Three things: a current price list, project information and the images to be shared. If a WhatsApp line is wanted, Meta verification starts on day one because it can take several days. Before launch the behaviour exam, the isolation proof and the legal texts are completed.
In a thirty minute call, see the system working with your own price list. Set whatever trap you like for the consultant and we will run the data isolation proof in front of you. Write to us with your project name, unit count and sales channels.