Valco AI - Real Estate Intelligence Platform for Dubai

Dubai publishes its property market as open data: records of sales, mortgages and gifts, registered units and rent contracts, updated daily. The data is public but not ready to use. It arrives as large CSV files, with names in Arabic and English, text where a database wants a key, empty values written as the word "null", and dates written day first. And the questions that matter are not the ones the files answer. A file lists what was sold. Valco's staff need to tell a client where the market is heading, which properties suit them and what is worth buying.

Automated ETL pipelines ingest the Land Department's files as streams, one line at a time, so file size doesn't matter; they cleanse and enrich the records and bulk-load them into staging tables in SQL Server and MongoDB. Set-based SQL then builds a model of the market: lookup tables for areas, projects, buildings, property types, room counts and the nearest landmark, metro station and mall; fact tables for transactions, units and rent contracts that share those lookups, so a building's sale prices and its rents sit side by side. On that model run time-series forecasting, clustering, anomaly detection and predictive scoring, written in Python on PyTorch, TensorFlow and Keras and including recurrent neural networks. Staff talk to the platform in plain language: it finds the best options for a client and answers with market insight in context. A Blazor dashboard presents the market intelligence and investment scores, and the platform produces the market analysis, segmentation and forecasting reports staff give to clients. Momentaj also licensed its DevGuardian AI platform to Valco AI and customised it for the company's internal development team.

Delivered to Valco Properties: the data pipelines, the market model, the forecasting and scoring models, the conversational recommendation system and the dashboard. The project is the template for how we approach applied machine learning: the data modelled first, models measured against it, and a conversational layer over a database rather than over a pile of documents.