SiteSense AI

Autonomous spatial inspection & temporal memory agent for civil infrastructure using dual multimodal vision, SurrealDB spatial graph memory, Clerk B2B RBAC, and Era by Eon guardrails.

Comment

Inspiration

What it does

How we built it

Challenges we ran into

🍁 Inspiration

Across Canada and North America, aging commercial real estate, parking garages, and high-rise foundations suffer from severe sub-grade environmental stress:

  • $3.4B+ Annual Damage: Freeze-thaw cycles, groundwater penetration, and micro-shifts degrade deep concrete foundations undetected.
  • Unmonitored Shear Fractures: Hairline micro-cracks quietly expand into catastrophic structural shear failures under hydrostatic pressure surges.
  • Regulatory Bottlenecks: Compliance with Ontario Building Code (OBC Part 4 - Structural Design & CSA A23.3) requires licensed Professional Engineers (P.Eng) to conduct manual site visits, estimate repair costs on clipboards, and draft offline work orders.

SiteSense AI bridges this critical gap as an autonomous spatial inspection and historical temporal memory agent that sees, measures, remembers, and quotes structural remediation with cryptographic P.Eng authority.

🏗️ What It Does

  • Autonomous Spatial Visual Grounding: Ingests live optical scans, CAD layouts, and point clouds to detect sub-millimeter shear fractures, concrete spalling, calcite efflorescence, and moisture intrusion.
  • SurrealDB Temporal Memory: Remembers spatial fracture progression over time (building ➔ floor ➔ zone ➔ crack), automatically calculating structural growth velocity (e.g., +800% widening under hydrostatic pressure).
  • Dual-Engine Multimodal Inspection: Combines ultra-low latency visual inference with zero-downtime hot standby failover.
  • Era by Eon Guardrails & Benchmarks: Enforces deterministic OBC Part 4 structural design validation, keeping hallucinations under 0.6% and compliance at 99.4%.
  • Instant Engineering Work Orders & RSMeans Costing: Auto-quotes union labor repair costs in CAD and issues certified field dispatch tickets.
  • Expo Mobile Companion: Provides sub-grade offline data queueing via WatermelonDB for connectivity-dead parking basements (P2/P3).

⚡ How We Built It (Sponsor Integrations)

  • SurrealDB 2.0 (Spatial Hierarchy & Temporal Graph): Preserves multi-dimensional site memory using directed graph relations (RELATE building->contains->floor->contains->zone->contains->crack). Evaluates historical deltas and fires reactive DEFINE EVENT critical_shear alerts for automated contractor dispatch.
  • Clerk (Enterprise B2B RBAC & PEO Authority): Multi-tenant organization boundaries separating General Contractors, Insurance Adjusters, and Lead Forensic Inspectors (P.Eng) with cryptographic sign-off authority (PEO Lic #100849204).
  • Nebius AI Studio (Primary Multimodal Engine): High-throughput visual inspection powered by Qwen/Qwen2-VL-72B-Instruct achieving 742 ms inference latency.
  • Hugging Face (Hot Standby Failover): Automated zero-downtime fallback to meta-llama/Llama-3.2-11B-Vision-Instruct via serverless inference router.
  • Era by Eon (Continuous Benchmarking): Automated evaluation suite testing against 100+ Ontario Building Code structural edge cases with 99.4% pass rates.
  • Expo SDK 51 (Field Mobile Companion): Offline-first local SQLite/WatermelonDB storage with over-the-air EAS updates for subterranean ## Accomplishments that we're proud of

What we learned

What's next for SiteSense AI

Built With

Share this project:

Updates

Submission history