Real Estate Club Dubai: a property platform built on real data

Dubai buyers piece together answers from sales material

International buyers, and people buying off-plan (before the building is finished), get most of their information from developer brochures and agency listings. Handover dates, fees and the quality of service providers are hard to check on your own. The official data sits in government registries that most people never open.

An AI assistant is only useful if it can't make things up

We wanted an assistant that could answer buyer questions and point people to real providers. A general chatbot would invent company names, fees and dates. Ours had to answer only from our own data and show its sources. It also had to stay cheap to run and hard to abuse on a public page.

One Laravel app with well-organised data at its core

The platform runs on Laravel 12 with Blade, Livewire 4 and Alpine.js. Listings, categories, developers, projects, areas, rankings, reports and articles are each stored as their own records. So each one gets its own page, and the other features can look it up. Background jobs run on Horizon and Redis (the tools that run tasks behind the scenes). Scheduled jobs sync data, send digests and run health checks.

Registry data and rankings you can check

A small connector reads the Dubai Land Department open-data service. A sync job runs twice a week and saves a snapshot of each project. The handover tracker compares official building progress and expected dates with each developer's own timeline. When the registry reading changes, it emails people who are watching that building. Rankings use a published formula with 6 to 8 weighted public criteria, and every score has a source. Paid status is left out of the calculation.

An AI advisor that looks things up instead of guessing

Ask RECD is built on Prism PHP, so we can switch AI providers. The default is OpenAI gpt-4o-mini. The model can't answer freely. It uses eight tools that search our own database: content search, directory and ranking lookups, project assets, handover status, glossary, market data, calculators and sending an inquiry. Each tool returns source links, so the answer can cite them. Replies appear word by word as they are written. Every turn is logged with its token use (how much text the AI processed) and cost. Before any AI call, a guard layer runs honeypot, bot and signed-token checks to stop bots and fake requests. On top of that there are limits per IP address, a daily cap for the whole site and a daily spend limit.