Skip to main content

Smart City Use Cases

Six production-grade deployments — all with measured KPIs, conformal guarantees, and EU AI Act evidence binding.

Adaptive traffic corridor (Tallinn pilot model)

Scenario: 18 km arterial with 34 signalised intersections, 220k daily vehicles.

Models: TrafficFlowOptimizer + IncidentDetector + MultimodalRoutingEngine

Measured KPIs
  • • Average travel time ↓ 18.4% (conformal 90% band: [15.1%, 22.0%])
  • • Signal-induced idling ↓ 27% → CO₂-eq savings 1,340 t/year
  • • Incident detection time p95 ↓ from 4m12s to 38s

Regulatory alignment: CSRD ESRS E1 (scope-1 road), ISO 37120 11.3/11.4, EU ITS Directive 2010/40/EU

Flood early-warning — Rhône basin (class HR-A)

Scenario: Pluvial + fluvial risk, 2,100 km² catchment, 17 gauging stations, NWP ensemble.

Models: FloodRiskPredictor ⚠️ (Annex III) + InfrastructureHealthMonitor + CityBrainOrchestrator ⚠️

Measured KPIs
  • • Lead time vs METEO baseline: +142 min (median)
  • • False-alarm ratio ↓ 0.21 → 0.08 with conformal calibration α=0.05
  • • HITL override rate 6.4% — all decisions Merkle-anchored

Regulatory alignment: EU AI Act Art.14 HITL gate · GDPR Art.89 · EU Floods Directive 2007/60/EC

Peak-shaving + DR — Baltic consortium (BP-106 DT federation)

Scenario: Tallinn + Riga + Vilnius + Helsinki — 4 TSO zones, 8.3 GW peak demand, 2,800 DR assets.

Models: EnergyGridOptimizer + DemandResponseAI + BatteryStorageOptimizer + CityDigitalTwinFederation

Measured KPIs
  • • Consortium peak ↓ 6.1% (€48M/yr avoided CapEx, aggregated)
  • • FedAvg round latency p99: 2.4s with Gaussian DP(ε=0.8, δ=1e-5)
  • • No raw telemetry leaves city boundary — GDPR Art.89 proof binding

Regulatory alignment: EU EED 2023/1791 · DP accounting log · NIS2 Art.21 evidence chain

ISO 37120 → CSRD ESRS auto-reporting

Scenario: Mid-size European city (320k inhabitants) → first full-scope CSRD FY2026 report.

Models: ESGIntelligence + EnergyGridOptimizer (consumption) + WaterDemandForecaster + WasteRouteOptimizer

Measured KPIs
  • • Manual effort ↓ from 2,200 h/year to 380 h/year (-82.7%)
  • • Evidence-trace coverage: 100% of E1/E2/E3/E5 datapoints Merkle-anchored
  • • EFRAG XBRL validation: 0 errors on first submission

Regulatory alignment: CSRD Directive 2022/2464 · ESRS E1/E2/E3/E5/S1 · ISO 37120:2018

Sovereign air-gapped city — Eastern EU capital

Scenario: 35-model kit deployed on RPi4 CM4 cluster, zero outbound egress, BLP MAC enforced.

Models: SovereignManifestValidator + full 35-model SC suite (offline conformal calibration)

Measured KPIs
  • • Cold-boot to first inference: 47 s
  • • Dilithium-3 manifest verification p95: 11 ms
  • • BLP MAC violation count in 90-day audit: 0

Regulatory alignment: BP-105a claim 4 · ANSSI SecNumCloud-compatible · eIDAS QES optional

Multi-agency emergency dispatch (City Brain)

Scenario: Heatwave + air-quality event, 1.2M inhabitants, 3 agencies (fire / medical / civil protection).

Models: CityBrainOrchestrator ⚠️ + EmergencyResponseCoordinator ⚠️ + AirQualityForecaster + PublicHealthSurveillance

Measured KPIs
  • • Mean dispatch latency ↓ 31% vs status-quo
  • • Cross-agency handoff errors ↓ 54%
  • • 100% of HR-A decisions HITL-gated, Art.13 rationale logged

Regulatory alignment: EU AI Act Annex III · Art.9 production gate · Art.14 HITL