Inspiration PropertyIQ was born out of a critical engineering realization: traditional building energy auditing and structural assessment methods are fundamentally bottlenecked by slow, fragmented, manual data collection. Our vision was to elevate AI from a conversational novelty to a true, domain-expert coworker capable of translating multi-modal field data into authoritative engineering artifacts. We sought to bridge the gap between abstract computer vision and strict physical reality, transforming a brief smartphone capture into an exhaustive, audit-grade analytical engine. What it does PropertyIQ transforms a rapid mobile spatial capture into a highly resolved Building Information Model (BIM) paired with comprehensive local law energy compliance profiles—executing a full analysis in under 3 seconds. Operating via a standard smartphone interface, the platform captures high-frequency depth maps and pairs them with external thermal sensor arrays to automatically identify structural elements, calculate complex heating and cooling loads, and detect thermodynamic anomalies. The resulting outputs directly automate deep technical reporting, generating localized compliance profiles such as NYC Local Law 97 metrics and ASHRAE validation frameworks. How we built it The technical foundation of PropertyIQ integrates rigorous physical models with an advanced multi-agent AI system: • The On-Device Stack: We built an iOS application utilizing the iPhone 15 Pro's LiDAR sensor to capture time-of-flight point clouds at 30 Hz, utilizing the A17 Pro Neural Engine to run local, privacy-preserving PointNet++ neural layers. • The AI Orchestrator System: Field telemetry feeds into our cloud-native FastAPI backend, where an Orchestration Controller dynamically manages specialized computational agents. These include a Thermal Analysis Agent for heat flux estimation, a Geometric Agent for boundary segmentation, and dedicated Compliance and Valuation Agents. • The Surrogate Moat: To deliver rapid results without cloud latency stalls, we engineered a Hierarchical Self-Correcting Multi-Fidelity Optimization Engine (HSMOE) that morphs dynamically between high-fidelity physics operators, fast Reduced-Order Models (ROMs), and Gaussian Process emulators based on active data certainty. Challenges we ran into Our primary hurdles centered on engineering-grade quality assurance and strict standards verification. While standard generative models can approximate visual environments, engineering utilities require strict boundary constraints. We encountered severe optimization instabilities and oscillatory shifting when transitioning between varying model fidelities. Furthermore, ensuring that autonomous agent deductions perfectly satisfied conservation of energy equations and rigorous building codes—such as ASHRAE 90.1—required building an entirely separate physics-informed verification layer to prune invalid mathematical inferences. Accomplishments that we're proud of We successfully decoupled deep building intelligence from manual consulting latency, transforming multi-day field processes into near-instantaneous, 3-second analytical outputs. We are proud to have engineered a fully realized, multi-agent AI ecosystem that guarantees strict adherence to core thermodynamic principles rather than statistical approximations. In lab-bench validation pathways, our system achieved a 95% accuracy rate relative to manual engineering appraisals while simultaneously unlocking significant energy cost reductions and localized carbon compliance improvements across monitored property portfolios. What we learned We discovered that the key to unlocking true AI utility in industrial fields lies in grounding neural networks inside classical mathematical frameworks. Integrating spatial and thermal data streams using Cross-Modal Attention layers yielded vast accuracy gains over single-source models. Most importantly, we learned that treating uncertainty as a core, quantifiable gradient allows the system to implement intelligent, hysteresis-damped control—ensuring that computational models remain highly stable and mathematically dependable even when processing imperfect or noisy real-world data. What's next for PropertyIQ Our roadmap is structured to drive national scalability and enterprise infrastructure integration. We are expanding our geometric models to ingest national-scale building archetypes and regional climate datasets. Architecturally, we are finalizing a secure, decentralized learning framework to allow multi-site property data to continuously refine global energy prediction models without violating enterprise privacy boundaries. PropertyIQ is evolving to become the unified, real-time operating fabric for the smart real estate asset management lifecycle.


Built With

Share this project:

Updates