SITE_INTELLIGENCE_PLATFORM
Site IntelligencePlatform forSchedule & RiskOptimization
Continuous fusion of vision, IoT/sensor feeds, and BIM/IFC into a live digital twin with GenAI copilots and RL optimizers for construction excellence.
OpenBIM/IFC
OpenDroneMap
Detectron2
OR-Tools
ORB-SLAM3
SCHEDULE_OVERRUN_%
20%
Current project delays
BUDGET_OVERRUN_%
18%
Cost overruns
REWORK_VARIANCE_%
22%
Quality issues
CONSTRUCTION_PROJECT_CHALLENGES
Large capital projects face fragmented ground truth
across RFIs, submittals, daily logs, photos, PDFs, drones, sensors, and BIM models
01
Schedule Overruns
Large capital projects routinely run 10–30% over schedule due to fragmented ground truth on site progress
IMPACT
10-30%
02
Budget Overruns
Projects run 15–25% over budget because planners rely on lagging updates and quantity takeoffs drift from reality
IMPACT
15-25%
03
Fragmented Data
Site progress, safety, quality, and supply data fragmented across RFIs, submittals, daily logs, photos, PDFs, drones, sensors
IMPACT
Data silos
04
Materials Quality Tracking
Materials quality (concrete, asphalt, steel) tracked in spreadsheets; test reports land weeks late
IMPACT
Weeks delay
Site Intelligence Platform
Continuous Fusion
Vision, IoT/sensor feeds, BIM/IFC
Live Digital Twin
BIM ↔ Reality alignment
GenAI Copilots
RFIs/submittals/claims
RL Optimizers
Equipment & crew dispatch
REAL, OPEN ANCHORS (AUDITABLE & MAINTAINABLE)
Real-World Foundation
01
OpenBIM / IFC
IFC4/IFC4x3 schemas; IfcOpenShell & BlenderBIM for reading/writing IFC and linking geometry to cost codes/WBS
02
Point clouds & photogrammetry
OpenDroneMap / COLMAP for photogrammetry, Open3D + PDAL for registration (ICP), meshing, clash/progress checks
03
Computer vision
Detectron2 / YOLOv8 for object/instance detection (PPE, equipment, materials), Segment-Anything/SAM2 for general segmentation
04
SLAM / tracking
ORB-SLAM3 (indoor/AR walk-throughs), AprilTags/ArUco for geo-anchoring
05
Scheduling & optimization
OR-Tools for routing/dispatch; NetworkX for precedence graphs; PyMC / Prophet / TFT for schedule risk forecasting
06
Structural/asset models
OpenSees (structural response), EnergyPlus/OpenStudio (building energy—useful for commissioning & OPR verification)
07
Sustainability/LCA
openLCA (methodology integration) and IFC property sets for EPDs (we do not redistribute databases; we integrate client-licensed sources)
Everything above is open, scriptable, and suitable for on-prem or VPC deployment
CONSTRUCTION_ARCHITECTURE
End-to-End Architecture
01
Field Data
360° cams
drones
phone photos
BLE/UWB tags
GPS
crane telematics
Concrete/asphalt batch data
NDT sensors
Weather & logistics feeds
↓
02
Data Fabric & Ingestion
Time-series ETL
media ingest
IFC import
Dedup/QA
privacy scrubbing
georegistration
↓
03
Site Digital Twin (BIM ↔ Reality)
BIM/IFC + point clouds/meshes
Pose/geo alignment
progress deltas
↓
04
AI Modules
Perception AI (Object/segm, Quantity takeoff, As-built deltas)
Schedule/Risk AI (Look-ahead sim, Probabilistic dates, RL crew/crane dispatch)
Safety AI (PPE & zone watch, Proximity alerts, Leading indicators)
Materials QC AI (Mix/lot traceability, QC test prediction, NDT/strength curves)
↓
05
Apps & Copilots
Planner web console
Superintendent mobile
QC lab portal
GenAI RFI/Submittal/Claim copilot (RAG on project docs)
MODULE_A
Perception & Progress
CV + Point Clouds
Key Features
1
Reality capture: Daily 360° walks, drone flights, and phone uploads auto-ingest
2
Registration: Open3D/ICP aligns new point clouds to the BIM coordinate frame; error surfaces flag misplacements
3
Semantic detection: Structural (columns, beams, decks, rebar, PT tendons), MEP (ducts, cable trays, equipment), sitework (rebar mats, rebar cover chairs, embeds)
4
Logistics: pallets, rebar bundles, pipe spools, man-lifts, excavators, tower crane hook state
5
Compliance: PPE (helmets, vests, harnesses), exclusion zones
6
Progress deltas: For each BIM element (GUID), compute % complete from geometric coverage and visibility confidence
7
Automated quantity takeoff: Volumetric comparisons between IFC geometry and reconstructed meshes; variance logged to cost work packages
Automated quantity takeoff with volumetric comparisons between IFC geometry and reconstructed meshes
MODULE_B
Schedule, Risk & Dispatch
Probabilistic + RL
Capabilities
1
Probabilistic schedule: Convert precedence network to a stochastic graph; learn activity duration distributions from historical + live productivity
2
Transformer/TFT models forecast crew productivity by activity (weather, congestion, learning curve), producing P-50/P-80 dates
3
What-if simulator: Adjust crews/equipment; simulate knock-on effects; surface critical chains (not just critical path) that drive slippage under uncertainty
4
RL dispatch (equipment & crews): State (tasks ready, locations, access, crane occupancy, delivery windows, weather), Action (assign crew/equipment, set sequence, crane timeslots, laydown area usage), Reward (negative makespan + penalties for congestion, rework, safety conflicts)
5
Output: hour-by-hour look-ahead with resource conflicts resolved and crane "slotting" recommendations
RL dispatch provides hour-by-hour look-ahead with resource conflicts resolved and crane slotting recommendations
MODULE_C
Safety Intelligence
Key Features
1
PPE & posture tracking with real-time alerts at entry points and high-risk zones (edge inference on site gateways)
2
Proximity risk: track workers vs heavy equipment trajectories; near-miss detection (speed, direction, separation)
3
Leading indicators: fuse housekeeping (clutter segmentation), access control violations, and weather/wind thresholds into a Safety Risk Index per zone
4
Auto-generate toolbox-talk topics for next shift based on safety risk patterns
Safety Risk Index per zone with auto-generated toolbox-talk topics for proactive safety management
MODULE_D
Materials Quality & Traceability
Concrete
Ingest batch tickets (slump, w/cm, admixtures), truck-to-pour times, embedded maturity sensors
Predict 28-day strength with uncertainty
Flag early formwork removal opportunities with safety margins
Steel
Heat/lot traceability linked to IFC elements
Weld/NDT results mapped to isometrics & spools
Auto-generate punchlists
Asphalt / Soils
Nuclear gauge & density tests correlated to mix design and compaction passes
Alert on failing compaction trends before paving windows close
Supplier scorecards: on-time, conformance, rework rates feed a sourcing KPI
MODULE_E
Claims, RFIs, Submittals
GenAI + Retrieval
RAG Corpus
Drawings, specs, method statements, RFIs, submittals, emails, meeting minutes, change orders → chunked, indexed
Drawings
Specs
Method statements
RFIs
Submittals
Emails
Meeting minutes
Change orders
Copilot Skills
Draft RFI answers citing sheet/detail paragraphs
Assemble submittal cover sheets referencing spec sections
Build time-impact analysis summaries from schedule deltas + weather + access logs
Create nonconformance reports with evidence (photos/point-cloud slices)
Guardrails: every response must include citations to governed docs; no free-text assertions
MODULE_F
Sustainability & Embodied Carbon
Key Features
1
Attach EPD data to IFC materials; compute element-level EPD rollups and project EPD totals
2
Recommend substitutions (e.g., SCMs in concrete) with modeled impact on strength/maturity curves and schedule risk
EPD integration with element-level rollups and project totals for comprehensive sustainability tracking
Web & Mobile Apps
Field-grade UX
Planner Web Console
BIM/IFC viewer with as-built overlays; progress heatmaps by zone/WBS; probabilistic Gantt (P-curves); crane timeline
"Why late?" explainer surfacing blocking predecessors, access conflicts, or material shortages
Sustainability dashboard (EPD rollups, deltas to targets)
Superintendent Mobile (offline-first)
Scan an element (QR/ArUco) → see status, punch items, latest photos/point-cloud slice
Approve AI-suggested look-ahead or override with notes
Safety alerts (wind + crane, proximity, exclusion breaches)
QC/Materials Portal
Batch tickets paired to pours/spreads; live maturity curves and predicted strip times
Test scheduling & results; weld/NDT map
GenAI Copilot
"Draft an RFI for clash at Grid C-7, Level 04; cite architectural & structural drawings."
"Build a time-impact summary for the façade delay, include weather days and access log extracts."
Data, Deployment & MLOps
| LAYER | TECH & PRACTICE |
|---|---|
Ingestion | S3/Blob for media; Kafka for events; IFC imports via IfcOpenShell |
Storage | Delta Lake (media manifests + metrics), PostGIS for geo-queries |
Feature store | Feast (per-activity productivity, weather, zone congestion) |
Training | PyTorch Lightning/Ray for vision & forecasting; OR-Tools for baselines |
Serving | Triton for CV; FastAPI for schedule/RL & copilot endpoints |
Edge | Jetson-class nodes on site gateways; model quantization (INT8) |
CI/CD | GitOps + model registry (MLflow); canary deploys; drift checks |
Security | RBAC, row-level permissions per contractor/sub; audit logs |
Interop | Exports to Primavera P6/MSP (XER/XML), Procore/ACONEX, COBie/IFC |
KPIs & Measurable Impact
Target bands you can defend
| DOMAIN | KPI | TYPICAL IMPROVEMENT BAND |
|---|---|---|
Schedule | Critical-chain reliability | +10–20 pts (P-80 adherence) |
Cost | Rework/variance vs WBS | −15–25% |
Productivity | Crew output variability | −10–18% std-dev |
Safety | Recordable incidents | −20–35% (via leading indicators) |
Materials | QC lead-time & waste | −15–30% (early strength + fewer retests) |
Claims | Cycle time to assemble TIA | −60–80% |
Sustainability | Embodied-carbon variance vs target | −10–20% |
Case Snapshots (Anonymized)
Patterned on real deployments
01
High-rise mixed-use (urban)
360° walks + drone weekly
RL crane slotting
28% hook conflicts reduced
Façade access conflicts
3 weeks earlier identification
Strip times advanced
14% of slabs using maturity
Progress auto-extract matched to WBS; RL crane slotting reduced hook conflicts; façade access conflicts identified earlier than baseline; strip times advanced safely using maturity.
02
Industrial plant (EPC)
Point-cloud vs BIM alignment
Pipe spool mis-supports
Caught before hydro
Weld/NDT map
~2% scrap prevented
Claims copilot
5 days to same day
Point-cloud vs BIM alignment caught pipe spool mis-supports before hydro; weld/NDT map prevented scrap; claims copilot cut RFI turnaround from 5 days to same day.
03
Civil package (roads/bridges)
Drone photogrammetry
Earthwork quantities
Reconciled daily
Compaction trend alerts
Reduced rework strips
Weather-risk look-ahead
Kept P-80
Drone photogrammetry → earthwork quantities reconciled daily; compaction trend alerts reduced rework strips; weather-risk look-ahead shifted crews to protected work, keeping P-80.
Implementation Roadmap
Fastest path to value
01
Weeks 0–2 — Connect Data
IFC import
photo/drone ingest
doc crawl for RAG
baseline schedule import
02
Weeks 2–6 — Perception & Schedule
Perception models tuned
progress & quantity deltas
probabilistic schedule
initial dashboards
03
Weeks 6–10 — RL & QC Pipeline
RL dispatch pilot
QC/maturity pipeline
concrete pours
safety leading indicators
04
Weeks 10–14 — Copilot & Claims
Copilot for RFIs/submittals
claims/TIA builder
sustainability EPD rollups
05
>14 weeks — Scale & Integrate
Scale to all zones/trades
integrate to P6/Procore workflows
add commissioning/OPR checks
Risks & Mitigations
Risk
CV false positives/negatives on cluttered sites
Mitigation
Multi-sensor fusion (360 + drone + phone); SAM-assisted labeling; human-in-the-loop review for low-confidence detections
Risk
Mis-registration of scans to BIM
Mitigation
Robust ICP with control points; QR/AprilTags; tolerance thresholds and manual nudge tools
Risk
Planner resistance to RL outputs
Mitigation
Explanations + side-by-side what-if; allow partial adoption (advisory mode)
Risk
Data ownership & privacy
Mitigation
Clear data contracts; on-prem/VPC option; per-contractor RBAC
Risk
EPD/licensing constraints
Mitigation
Use client-licensed LCA datasets; store only references in IFC properties
Continuously learning system
Tensorblue turns jobsites and materials plants into a continuously learning system. By aligning BIM/IFC with reality capture and decision AI (GenAI for paperwork, RL for dispatch, predictive models for schedule/safety/QC), we remove latency between plan and ground truth.
Result: fewer surprises, fewer claims, safer shifts, and smoother handover — with evidence attached to every decision