Propwise

The problem

A multi-tenant real estate platform needed strict data separation between agencies, fast bilingual search, reliable background work and one place to publish to the main UAE property portals.

What I did

• Enforced tenant isolation with PostgreSQL Row-Level Security, so a missing filter in application code can never leak data across tenants.
• Built the async foundation on a Postgres-native job queue (pg-boss): concurrency, deduplication, retries with backoff and reconciliation sweepers so no job is silently dropped.
• Replaced a slow multi-table ILIKE search with a denormalized index and async indexer, plus full-text/trigram routing over GIN indexes and Arabic/mixed-script normalization.
• Designed two-layer RBAC (endpoint guards + per-record flags) with a 5-minute permission cache and instant invalidation.
• Built tamper-resistant audit logging: triggers on 35 tables record every change with before/after values, user, session and IP.
• Built syndication to Property Finder, Bayut and dubizzle with a 9-state sync state machine, plus automated lead capture and a lead-distribution and escalation engine.

The result

• Median search ~800 ms → under 100 ms
• ~70% fewer authorization lookups
• Customer data separated by design
• Listings on 3 portals from one place

Technology

• PostgreSQL
• Row-Level Security
• pg-boss
• GIN / trigram indexes
• TypeScript
• Node.js
• React