RealtyDesk: Building a 360° Property Ecosystem with Sub-Second Search Across 25,000 Listings

The client had spent six months and a significant portion of their seed funding with a previous development agency before coming to us. The previous vendor had delivered a partially functional listing portal — static search, no mobile app, no agent verification, and a backend architecture that the client's new CTO described as "technically a website from 2014." The product had been demoed to investors with the promise of launching within 60 days. It wasn't close to launchable.

The real challenge was not rebuilding the portal — it was building the right thing quickly and well. The client's market insight was correct and defensible: property buyers in their target Southeast Asian market bounced between four or five fragmented platforms to research a property, verify an agent, source interior services, and arrange legal documentation. No single platform unified these touchpoints. But the opportunity window was closing: a well-funded competitor had announced a similar product at a regional property conference three months earlier.

The technical requirements were genuinely demanding. The search experience needed to return geo-filtered results across 25,000+ listings in under one second — not because it was a nice-to-have, but because internal A/B testing from comparable markets showed that search response times above 1.5 seconds caused a 40% drop in query completion. Agent and buyer verification needed to happen digitally, fast, and in a way that built trust without creating onboarding friction. And the AI recommendation engine needed to produce relevant suggestions from day one — before the platform had accumulated the behavioural data that typically trains recommendation systems.

We made four architectural decisions in week one that shaped the entire build. First, Elasticsearch at the core of property search — not a traditional database query. PostgreSQL handles data persistence; Elasticsearch handles every user-facing search, with geo-bounding, multi-filter, and full-text support indexed and optimised for sub-second response. This separation meant we could tune search performance independently from the transactional data layer.

Second, an eKYC integration for agent and buyer verification with a target of under 2 minutes for median completion. We evaluated three eKYC providers against document processing accuracy, API reliability, and local ID document coverage. The selected provider handled the target market's national ID and passport formats with over 97% first-attempt accuracy and an API SLA of 99.9% uptime. Verification is embedded in the onboarding flow, not tacked on as a separate step — reducing the psychological friction of the compliance requirement.

Third, a cold-start recommendation strategy. A pure collaborative filtering engine has nothing to learn from on day one. We seeded the recommendation engine with content-based similarity scores — property attributes, location, price band, property type — allowing the system to make relevant suggestions before any user behaviour data existed. As real behavioural signals accumulated (searches, saved properties, enquiries), the engine transitioned to hybrid recommendations over the first 6 weeks post-launch. The 3x engagement lift over browse-only was measured at the 6-week mark, not day one.

Fourth, a human-in-the-loop moderation system for listing quality. Automated photo quality scoring (blurry, insufficient coverage, misleading crops), duplicate detection, and price anomaly flagging handle the majority of quality issues. Borderline cases surface to a moderation queue staffed by the client's team. This two-layer approach kept listing quality at 98% without requiring a full-time moderation team for every submission.

RealtyDesk delivered the Angular web application, Flutter iOS and Android mobile apps, .NET microservices backend, Elasticsearch search cluster, and a React moderation and analytics dashboard across 22 weeks. The property search delivers sub-1-second results across 25,000+ live listings with filters across property type, geo-radius, price band, bedroom count, amenities, and agent rating — all applied simultaneously without search latency.

The agent marketplace module allows verified agents to create rich profiles, manage their listing portfolio, and respond to buyer enquiries through an in-app messaging system. Agent ratings aggregate from verified transaction completions — not self-reported or gaming-prone. Buyers see agent response time, listing accuracy score, and verified transaction count before initiating contact, creating transparency that increases enquiry-to-viewing conversion.

The interior services and legal consultation modules were delivered as marketplace extensions within the same app. Interior designers and property lawyers create service listings; buyers can shortlist and request quotes without leaving the platform. For the client, these modules created two new revenue streams — commission on interior service bookings and lead-generation fees to legal firms — that had not been in the original scope but were added in a 2-week sprint after the platform validated interest during beta testing.