Time Series Graph Neural Nets Traffic AI Simulation Urban Planning

Neural City Digital Twin for Traffic Forecasting and Planning

Created a digital twin that learns city-scale traffic dynamics from sensors, map topology, and events. The system predicts congestion 5 to 60 minutes ahead and estimates intervention impact before rollout, enabling planners to test policies with lower operational risk.

-27%Peak Congestion (Sim)
91%Forecast Reliability
4.3xPlanning Throughput
320Intersections Modeled

City Sensing to Policy Simulation Loop

The twin fuses stream data, learns spatiotemporal patterns, predicts pressure points, and evaluates policy candidates before deployment.

Sensors + Events -> Graph Encoder -> Forecast -> Simulator -> Intervention Ranking
Inputs
📷
Loops + Cameras Live Traffic Streams
Sensor Feed Fusion

Ingests loop detectors and camera counters to estimate flow, occupancy, and queue pressure in real time.

StreamingSensor Fusion
🗺️
Road Graph Topology + Lanes
Network Topology

Represents intersections and corridors as a graph with lane-level connectivity and turn constraints.

GraphMap Priors
Context
🌧️
Events Incidents + Weather
Event Context

Adds dynamic context from weather, incidents, and event schedules to anticipate non-stationary demand spikes.

WeatherIncidents
📅
Calendar Priors School + Holidays
Temporal Priors

Encodes day-of-week and special-day priors to stabilize long-range forecasts and reduce shift sensitivity.

SeasonalityPriors
Decision Loop
📈
Graph Forecaster 5-60 min Horizon
Spatiotemporal Forecasting

Graph model predicts near-term congestion propagation across connected corridors and bottleneck nodes.

GNNForecast
🧪
Policy Simulator Signal + Routing Plans
Intervention Sandbox

Simulates timing plans and routing policies to estimate delay, throughput, and spillback impact before rollout.

What-ifSimulation
🎯
Top-N Actions Expected Uplift
Action Ranking

Ranks candidate interventions by projected uplift with uncertainty-aware confidence scoring.

RankingImpact

Synthetic Congestion Map Snapshot

Low Load Moderate Load High Load Primary Corridor

Map visual is a dashboard-style abstraction used for rapid policy preview, not a geospatial rendering.

Interactive Plots

Congestion Forecast with Confidence Band

Predicted corridor load over 60 minutes

Line Band

Scroll to zoom · Drag to pan

Intervention Uplift by Strategy

Delay reduction and throughput gain

Mixed

Scroll to zoom · Drag to pan

Hotspot Risk by Intersection

Bubble size = queue spillback probability

Bubble

Scroll to zoom · Drag to pan

Incident Composition

Traffic disturbance source mix

Polar Area

Scroll to zoom · Drag to pan

Scenario Family Performance

Model resilience in event-heavy windows

Radar

Scroll to zoom · Drag to pan

Compute vs Accuracy Frontier

Deployment tradeoff across model sizes

Scatter

Scroll to zoom · Drag to pan

Business Impact and Delivery Scope

Problem Solved

City operations teams lack reliable forecasting and simulation tools to test interventions before field rollout.

What I Deliver

Digital twin architecture integrating traffic streams, graph forecasting, and intervention scenario simulation.

Expected Impact

Lower congestion peaks, faster planning cycles, and better evidence for policy decisions.

Hire Me for Digital Twin and Forecasting Systems

I can support urban mobility teams building predictive simulation and decision-support platforms.

MVP Delivery

Forecasting baseline and scenario simulator for one priority corridor or district.

Production Hardening

Data reliability checks, model drift controls, and stakeholder-facing decision dashboards.

Advisory + Build

Roadmap and implementation support for city-scale mobility intelligence programs.

Other Projects

Agent Reliability Lab

Observability and guardrails for tool-using agent systems.

Weather-Resilient Perception

Condition-aware autonomous perception with adaptive fusion.

Document Intelligence Copilot

Evidence-grounded QA over enterprise documentation.

Pedestrian Awareness

Audio-visual detection and motion awareness at the edge.