Proptech
| B2B SaaS | 3.5 Months | Beta Release
Asset & Property Management Platform
MY ROLE
Founding Product Designer
Vision Definition
Architecture Strategy
UX Advocator
STAKEHOLDERS
1 CTO (Founder)
2 Software engineers
1 MEP engineer
1 Product designer(Me)
IMPACT
Build an intelligent cross-role platform with 3D IoT raw data from end to end, improving fault localization speed by 35% and enabling AI-assisted decision-making across the enterprise property management and analysis lifecycle.
MISSON & PROBLEM
Clear Mountain Capital is a New York–based PropTech investment firm founded in 2014, with over $100M invested in companies like Docker, Uber, Bitly, and Optimizely. Its mission is to maximize returns through intelligent, data-driven property operations.
Commissioned to deliver a next-generation real estate operations platform, I led the design of a scalable SaaS dashboard unifying mixed reality BIM visualization, IoT telemetry, and facility management workflows.
At the outset of the project, the scope was vague. Our client stated they needed “predictive 3D intelligence” but lacked a specific data structure or a focused platform.
No Shared Visibility
Administrators waste 33% of their time navigating disconnected systems.
IoT ≠ Intelligence
85% of IoT data remains unused without context, visualization, and actionable insight.
Reactive Maintenance
Audits show reactive maintenance costs 3–10× more annually than preventative strategies.
FINAL DELIVIABLE
Across Hesta’s full maintenance lifecycle, I shipped 3 main flows that connects spatial awareness, intelligent task routing, and transparent AI insights, reducing operational friction and helping teams act faster with more confidence.
Spatial-to-Issue Detection
Accountability Workflow
Analytics & AI Reliability
Within the beta launched(≈10-12 weeks)
↑ 72%
AI Adoption
Increased 65–72% as transparency and data lineage improved.
Task creation was 65% faster, duplicate requests decreased by 52%.
↑ 40%
Operational Efficiency
$400K
Annual Savings
A 20–25% reduction in operational costs and $150K–$400K in annual savings for mid- to large-scale buildings.
USER RESEARCH
Diagnosing system complexity to accelerate MVP alignment
To align engineering, product, and operations around a unified system vision, I created this diagram mapping real-time data infrastructure to business-critical modules.
Defined the role-based access features and development priority
With no researchers and product manager on the team, I led early product scoping, interviewing 35+ users and prioritizing modules that improved resolution speed, role handoff, and traceability, while framing ESG and lifecycle tools as scalable extensions for long-term value.
Tenants & Occupants
- Occupy multi-family units
- Submit maintenance requests
- Provide feedback
- Seek mobile access
“ We were exhausted from reporting issues and tracking status across 10+ applications.
Property Managers
- Manager 14+ parks & retail
- Coordinate vendors
- Track portfolio data
- Optimize budgets
“ Observing and deciding whether to expand response took up 60% of our time.
Maintenance Vendors
- Cover 20+ properties
- Receive 50+ maintenance requests
- Delegate and monitor
- Report status manually
“ We need go from alarm to work order with one click, and be able to accurately land on 3D.
Admins & Analysts
- Analyze performance metrics
- Ensure compliance across portfolio
- Generate performance reports
- Identify and predict risk
“ Our ESG efforts are blind and slow, with data analysis buried in 20+ different places.
Avoid a one-size-fits-all information architecture diagram
Then, I mapped their specific actions, handoffs, and pain points onto one unified place, which visualizes how maintenance tickets, smart alerts, and SLA tracking flow across roles, ensuring that nothing falls through the cracks.
Working closely with engineers, I enforced a “one-tap from alert to action” rule, so the IA adapts dynamically as issues escalate, rather than forcing users to dig through sections.
How might I design a 3D hologram platform that help multi-stakeholders locate property issues directly, create issue orders faster, and analyze system health more easily?
IDEATION
Sketch A: Immediate perception for deep analysis and prediction
They love it
Supports situational awareness at a glance.
Simultaneously see open issues and system health.
Reduce context switching by combining event sourcing and KPI snapshots into a single view.
High information density works well in command centers and duty rooms with large displays.
Could be better
Brings cognitive overload and visual crowding on devices like 15–16” laptops, iPad, tablet.
New users may face a steeper learning curve to parse dense data quickly.
Not optimized for long-term daily operations where scrolling and gradual drill-down are more natural.
Sketch B: Gradual contextual understanding for quick task execution
They love it
Provides a clear reading order: locate the space, view work orders, and review KPIs/trends.
Modules can be extended or rearranged as the system grows.
Lower cognitive load, easier for daily operators to adapt and maintain workflows.
Could be better
Requires scrolling and section switching, which introduces minor context switching overhead.
Less effective for high-severity emergencies, as managers need quicker “all-in-one” snapshots.
Macro-level insights are less immediate compared to the dense overview of Ideation A.
Solution C: Unification system with cross-role task synchronization
Unified data model syncs 2D/3D and IoT in real time, cutting front-end complexity.
Reusable APIs accelerate delivery and maintain UX consistency.
An extensible data layer enables seamless ESG and predictive analytics integration.
In early testing, 17/22 participants preferred the solution C mode-switch layout for its “cleaner visual hierarchy” and “less cluttered workspace.”
Average task completion time for locating and responding to alerts dropped by 21%. (vs. scrolling or split-screen layouts).
KEY ITERATION
To keep this case concise, I’m showing just the two key iterations here. I’d be glad to share more details about the full iteration process and user feedback during my presentation.
Challenge 1:
Hybrid-rendering Logic Node System
Because our clients use heterogeneous BIM and 3D models, issue nodes were often misaligned, buried behind geometry, or visually lost under dynamic lighting. This meant the same “issue node” could appear floating, buried inside walls, or invisible, depending on lighting and depth composition.
• Unity-rendered twins had custom shaders and baked lighting, distorting icon color visibility.
• Revit models contained dense parametric metadata but inconsistent coordinate origins.
• IFC models were lightweight but lacked precise surface normals for spatial anchoring.
Solution:
I mapped failure points by categorizing visual errors (contrast loss), spatial errors (coordinate drift), and performance drops (frame rate dips). I also learned about color systems from brand designers and started exploring and creating new 0-1 design system.
Then, I partnered with engineers to log average Z-offset variance (in cm) and node visibility rate (%) per model type. We conducted cross-model validation sessions using sample buildings:
• Lighting test: varied HDRI skyboxes and internal emissive lights → observed node clarity.
• Depth test: applied different clipping planes and Z-bias → ensured consistent projection.
• Performance test: measured FPS stability under 1000+ concurrent nodes.
After receiving strong feedback from our field engineers and clients integrators, I built a Figma-to-Unity token bridge that standardized how issue nodes render across heterogeneous BIM (Revit/IFC) and Digital Twin models.
Each node property shape semantics, halo intensity, contrast tier, depth bias, occlusion logic, and LOD thresholds were tokenized in Figma, serialized into JSON schemas, and automatically parameterized in Unity.
Fault Localization
+45% Faster
Field users located faulty assets 41-45% faster, even in dense floor plans or low-light industrial spaces.
Modeling Development
-16% Code
Cut model-specific rendering code by 16%, consolidating all node variants under a single adaptive component.
Challenge 2:
Multimodal AI Chatbot Foundation
Our primary clients' users needed an AI assistant. Within time constraints, our develop lead proposed built an AI FAQ Bot prototype. So I came up with a preliminary solution:
“When I just need the energy consumption data, the FAQ on the right feels like too much information”
—— Property Manager
“The report is useful, but I need to know where the data came from IoT sensors or manual logs?”
—— Data Analyst
“Why isn’t there a cancel option? Sometimes I ask the bot to generate a report, then realize I missed a detail.”
—— Facilities Technician
Solution:
An early version in user interviews, we found that the information density and multiple layers of feedback caused distraction.
The subsequent version simplified the display of frequently asked questions, unified the metadata structure, and introduced role filtering, making the interface more focused, professional, and more in line with daily workflows.
Workflow Automation
Avg. 2min → 45s
Average task time dropped approximately 2.7 times faster, as users spent less time switching between chat, FAQ, and confirmation steps.
User Trust in Chatbot
↑ to 83%
Trust in AI-generated reports rose from 52% to 83%, driven by transparent metadata such as “last refreshed” and data-source visibility.
REFLECTION