Architecture & benchmarks

The demos are small, but they are built like production systems: reproducible data pipelines, typed APIs, a database, and a 3D client tuned for large datasets. Performance numbers below are measured in visitors' browsers, not estimated.

System overview

DATA SOURCES PIPELINES SERVICES & STORAGE 3D CLIENT

Per-demo pipeline

DemoInputProcessingDelivered asClient technique
City Digital TwinGIS footprints (synthetic or OSM), sensor registryExtrusion attributes; IoT simulator as a pure function of timeGeoJSON + SSE stream + snapshot / series APIOne batched Primitive for all buildings, per-instance colour updates, PolylineCollection for traffic
BIM–GISIFC4 / IFC2x3web-ifc (WebAssembly) parse → mesh extraction → simplification → georeference from IfcSite → semantic mappingIn-browser conversion; nothing uploadedGeometryInstance per IFC element (pickable by expressID), 4D filter by property-set dates
Asset ManagementAsset register, inspections, defectsValidated CRUD, workflow states, condition analyticsREST API + SQLite (PostGIS in production) + CSV exportProcedural 3D bridges with defects in structure-local coordinates
Underground & TunnelUtility network, tunnel alignment, monitoringDepth-aware 3D pipe geometry, ring schedule, settlement seriesGeoJSON / JSONGlobe translucency, underground camera, clearance analysis
Metro Line 3OSM alignment (viaduct + tunnel), stations, corridor buildings & roadsChainage & vertical profile, viaduct/pier/tunnel geometry, timetable-driven train simulationJSON / GeoJSONInstanced trains on a polyline-chainage model, follow camera, globe translucency for the tunnel section
Point CloudLAS/LAZ scans (synthetic by default)Classification, LOD tiling, DSM gridding, epoch differencing3D Tiles (.pnts) + DSM gridsStreaming LOD, eye-dome lighting, style expressions, volume from DSM
Urban PlanningParcels with zoning rules, buildingsRule engine (height, FAR, coverage, setback)GeoJSONInteractive massing, sun-position shadows, scenario compare

Datasets

Generated by the scripts in tools/ with fixed seeds, so every number is reproducible.

Loading…

Measured performance

Every demo page measures its own load time (navigation → interactive scene) and the average frame rate over the first seconds, then reports it anonymously. Automated test runs are excluded.

Takes about 2 minutes; demos load one after another in a hidden frame.
Loading…

From demo to production

What changes

  • IoT simulator → MQTT / Kafka ingestion with TimescaleDB or InfluxDB; the snapshot / series API stays the same.
  • SQLite → PostgreSQL + PostGIS, with spatial indexes and history tables.
  • Static 3D Tiles → object storage + CDN, or Cesium ion / self-hosted tiling for large scans and photogrammetry.
  • Synthetic city → OSM, cadastral data, Google Photorealistic 3D Tiles or customer surveys.

What we add

  • Authentication (OIDC / Keycloak), role-based access per layer and project.
  • Audit trail for inspections and work orders, file attachments, mobile field forms.
  • CI/CD, monitoring, backup, and performance budgets per dataset.
  • Integrations: ERP / CMMS, BMS / SCADA, document management, BI dashboards.