The Data & AI Behind Aumra Nova | Dubai Real Estate Intelligence

The data and models behind Aumra NovaAumra Nova runs four production systems on one feature store: a deterministic property score, a distressed-inventory sourcing pipeline, a retrieval-augmented advisor, and a lead qualification classifier. All of them read the same normalised records, and the scoring logic is published in full on our methodology page.

Pipeline

Ingestion. Developer and brokerage inventory feeds, direct seller submissions, Dubai Land Department transaction history, and our own advisory pipeline.
Extraction. Language models parse unstructured seller messages and brochures into structured fields. Every record keeps its raw source text for audit.
Deduplication. Units are matched on project, bedrooms, saleable area and price, so the same unit posted by several brokers is counted once.
Scoring. A deterministic weighted model produces the property score and tier from five signals.
Validation. Outlier prices and impossible unit sizes are rejected; aggregates are suppressed below minimum sample size.
Serving. Daily aggregate recomputation, continuous inventory refresh, and open dataset publication as HTML, JSON and CSV.

Retrieval-augmented advisory

The advisor answers only from verified records in our own database. It is constrained to retrieved inventory rather than free generation, so it cannot invent a project, a price or a payment plan. Where a record does not exist, it says so instead of estimating.

Last updated: 2026-08-31. Source: Aumra Nova live inventory and Dubai Land Department transaction records. Figures are recomputed daily.Home · Dubai properties · Off-market deals · Market data · Methodology · Why Aumra · Insights