HHD Real Estate AI Brief

Model built on Mask R-CNN base, trained on annotated images converted to COCO data format.

System detects lighting symbols, classifies light types, and extracts associated text annotations.

Target output: per-floor light count, fixture type, and power requirement estimate.

Benchmark built by 26 native Arabic speakers across 13 countries covering 12 everyday topics.

Strongest models score mid-90s on culturally appropriate reply selection but succeed on dialect production in only half of cases.

North African and Emirati dialects ranked among the most difficult for current models.

Round led by Coatue, with Blackstone, MGX, T. Rowe Price, and Sixth Street Growth among roughly 24 named participants.

Annualised run-rate revenue at $7bn, growing 80%; Lakebase agent database hit $100m run rate since June 2025 launch.

Company had already raised $20bn over the prior 20 months before this close.