Built at c0mpiled × Razorpay Hackathon
An AI-powered real estate decision system for the Dubai market — combining market intelligence, ROI analysis, and photorealistic 3D property exploration into a single investor workflow.
Dubai real estate is a global market where most investors never visit in person.
Investors rely on static images, fragmented data across portals, and WhatsApp voice notes from brokers.
Brokers lack shareable, high-quality assets and struggle to build trust remotely.
The gap: there's no platform that closes the loop between data → insight → experience → decision.
This is not a listing platform. Not a 3D tour tool. Not an ROI calculator.
It's a decision system — built on the premise that the right question in real estate isn't "show me the property" but "help me understand if this makes sense."
That framing drove every product decision made during the build.
A 4-layer system, each feeding into the next:
| Layer | Name | What It Does |
|---|---|---|
| 1 | Landing Platform | Property discovery and entry point |
| 2 | ROI Wizard | Yield, IRR, and cashflow analysis |
| 3 | Prop Pulse | Area trends, price benchmarks, supply pipeline |
| 4 | 3D World Viewer | Photorealistic Gaussian Splat exploration with spatial annotations |
Core flow: Discover → Analyze → Experience → Share
The deliberate design choice: every tool feeds into the 3D viewer, not away from it. The 3D experience is the convergence point, not a standalone feature.
AI isn't a layer on top — it's embedded in the decision loop:
- ROI Wizard: AI surfaces investment viability signals (yield, IRR, cashflow) from raw listing data, reducing manual calculation to near-zero
- Spatial Annotations: AI tags 3D environments with contextual markers — ROI projections, finish quality, view premiums — directly in the space
- WhatsApp-ready tours: AI generates 30-second cinematic clips from the 3D model, optimized for broker-to-investor sharing on mobile
The AI bet: In remote transactions, trust is visual. The highest-leverage use of AI here is not search or recommendations — it's generating shareable conviction.
1. Shareability as a core primitive, not an afterthought
Every output — ROI report, market snapshot, 3D tour — was designed to be one-tap shareable via WhatsApp. This was a deliberate constraint that shaped the entire UX. Brokers don't use portals; they use messaging. We built for that reality.
2. Workflow over features
Early temptation was to build each tool standalone. The PM decision was to force a connected flow — you can't skip straight to 3D without seeing the data. This raised friction slightly but dramatically improved decision quality per session.
3. Speed over perfection on the AI outputs
The 3D generation from imperfect photos was the hardest technical constraint. Rather than block on perfect quality, we optimized for "good enough to trust" — conviction in minutes, not days. A deliberate tradeoff to ship within hackathon constraints.
The problems that don't show up in a product brief:
- Input quality vs. output trust: Gaussian Splat quality degrades with sparse or low-quality photo sets. Users can't trust a tour that looks wrong. Defining the minimum viable photo set for acceptable output was a real design problem.
- Annotation accuracy: Spatial ROI annotations are only as good as the underlying data. If Prop Pulse data is stale, the annotation misleads. Surfacing data freshness signals in-product was a late but critical addition.
- Broker adoption friction: Generating the cinematic clip works — getting brokers to change the habit of manual WhatsApp pitches doesn't happen automatically. Distribution design is as hard as the feature itself.
Not vanity metrics. Decision quality metrics:
- Tool → 3D conversion rate — are users reaching the core product?
- Video shares per session — is the distribution primitive actually being used?
- Session depth — are users exploring beyond the first property?
- SaaS subscription for brokers (access to Prop Pulse + ROI Wizard)
- Per-property 3D scan fees (supply-side monetization)
- Enterprise / white-label for developers and agencies
If this were to move beyond hackathon scope, the next bets:
- AI-generated staging — let investors visualize furnished vs. empty scenarios
- Automated annotations at scale — move from manual tagging to model-driven spatial markup
- Portfolio-level analytics — an investor with 3 properties needs a different view than a first-time buyer
- Off-plan / developer support — the largest inventory segment in Dubai, currently underserved by 3D
| Platform | What It Does |
|---|---|
| Matterport | 3D tours (no data, no AI layer) |
| Property Finder | Listings (no 3D, no decision support) |
| Luma AI | 3D generation (no real estate context) |
This product sits at the intersection of all three — and is the only one where the output is a decision, not just a view.
- Dubai real estate market at record transaction volume
- Remote investing is now the default, not the exception
- 3D + AI infrastructure is finally viable at hackathon-level cost and speed
Built to explore how AI, spatial computing, and product design can reshape the real estate decision loop — from fragmented data and static images to a shareable, AI-augmented experience that closes deals remotely.
This was a hackathon build. The thinking behind it is production-grade.