DEV Community

Seyed Alireza Alhosseini
Seyed Alireza Alhosseini

Posted on

From Synthetic Citizens to Living Cities Building a Population Twin for the AI-Native City

What if a city could do more than monitor itself?

What if it could understand the people living inside it, simulate their collective behavior, test policies before deploying them, anticipate unintended consequences, and continuously learn from reality?

That is the direction behind Living City OS.

And one of the foundations for building it is not another dashboard, another IoT platform, or another traffic-management system.

It is a synthetic population that behaves like a living component of the city.

This is the idea behind my project:

Dubai Persona DAG

GitHub Repository


The Problem With Today's Smart Cities

Most Smart City architectures begin with physical infrastructure:

  • Sensors
  • Cameras
  • Buildings
  • Roads
  • Energy grids
  • Water networks
  • Traffic systems
  • Environmental monitoring

These systems generate enormous amounts of data.

But there is a fundamental problem:

A city is not just infrastructure.

A city is millions of decisions made by millions of people.

A new metro line changes mobility.

A change in fuel prices changes transportation choices.

A new visa policy changes population composition.

A heatwave changes electricity consumption.

A change in housing prices changes where people live.

A new business district changes commuting patterns.

These effects interact.

The city is therefore not a collection of independent systems.

It is a complex adaptive system.

Recent research on urban digital twins is moving in exactly this direction: from passive representations of infrastructure toward systems capable of simulation, intervention, and increasingly autonomous management. (arXiv)

But there is still a missing layer.

The Human Layer.


The Missing Twin: Population

A conventional Digital Twin might represent:

Roads
Buildings
Energy
Water
Climate
Traffic
Enter fullscreen mode Exit fullscreen mode

An AI-native city needs another layer:

People
Households
Jobs
Income
Mobility
Preferences
Consumption
Behavior
Enter fullscreen mode Exit fullscreen mode

This is where Dubai Persona DAG becomes interesting.

The repository generates 1,000,000 synthetic Dubai resident personas using a 48-dimensional probabilistic DAG, calibrated against official UAE data sources. (GitHub)

The goal is not to recreate real individuals.

It is to construct a statistically grounded synthetic population that can represent the structure and dependencies of a real urban population without exposing private individuals.

That distinction is critical.


From Persona Generator to Population Twin

Today, the architecture can be viewed as:

Demographic Variables
        ↓
Probabilistic DAG
        ↓
Synthetic Persona
Enter fullscreen mode Exit fullscreen mode

But this is only the beginning.

The next step is:

Synthetic Population
        ↓
Behavioral Models
        ↓
Mobility Models
        ↓
Economic Models
        ↓
Environmental Interaction
        ↓
City Simulation
Enter fullscreen mode Exit fullscreen mode

The synthetic citizen becomes an agent.

The population becomes a Population Twin.

And the Population Twin becomes part of the city's Digital Twin.


The Living City Architecture

My broader vision is a system I call:

Living City OS

The architecture can be organized into five layers.

                     LIVING CITY OS
                           │
              ┌────────────┴────────────┐
              │                         │
        PHYSICAL TWIN             POPULATION TWIN
              │                         │
      Roads / Buildings          Synthetic Citizens
      Energy / Water              Households
      Climate / Infrastructure    Behaviors
              │                         │
              └────────────┬────────────┘
                           ↓
                    CITY CAUSAL GRAPH
                           ↓
                    SIMULATION ENGINE
                           ↓
                      CITY BRAIN
                           ↓
                    POLICY / ACTION
                           ↓
                      REAL CITY
                           │
                           └──────→ DATA ──→ Feedback
Enter fullscreen mode Exit fullscreen mode

This creates something fundamentally different from a conventional Smart City.

Instead of:

Sense → Dashboard → Human Decision

we move toward:

Sense → Understand → Simulate → Predict → Decide → Act → Learn


City Genome

A useful abstraction for this architecture is the City Genome.

The city has multiple interacting genomes.

1. Physical Genome

The physical structure of the city:

  • Buildings
  • Roads
  • Bridges
  • Water networks
  • Power networks
  • Public spaces

2. Population Genome

The statistical structure of its inhabitants:

  • Demographics
  • Employment
  • Housing
  • Income
  • Mobility
  • Digital behavior
  • Consumption

3. Behavioral Genome

The patterns that emerge from interactions:

  • Where people move
  • When they move
  • What they consume
  • How they respond to prices
  • How they respond to policies
  • How they react to environmental conditions

Dubai Persona DAG provides an important foundation for the second layer.


Why a DAG?

A city is not a flat dataset.

Variables depend on one another.

For example:

Nationality
     ↓
Visa
     ↓
Employment
     ↓
Salary
     ↓
Housing
     ↓
Neighborhood
     ↓
Transportation
     ↓
Daily Mobility
Enter fullscreen mode Exit fullscreen mode

This is fundamentally different from generating random records independently.

The objective is to preserve dependencies.

That matters because urban behavior emerges from these dependencies.

A person doesn't choose transportation in isolation.

Transportation may depend on:

  • Income
  • Housing location
  • Employment location
  • Family structure
  • Vehicle ownership
  • Metro accessibility
  • Time constraints

A realistic simulation therefore needs to preserve the structure connecting these variables.


From Static Personas to Autonomous Citizens

The next evolution is to transform each synthetic persona into an agent.

Instead of:

{
  "income": 15000,
  "housing": "apartment",
  "transport": "car"
}
Enter fullscreen mode Exit fullscreen mode

we want:

Identity
    ↓
State
    ↓
Goals
    ↓
Constraints
    ↓
Preferences
    ↓
Environment
    ↓
Decision
    ↓
Action
    ↓
New State
Enter fullscreen mode Exit fullscreen mode

Now the citizen has a trajectory.

The agent wakes up.

It chooses transportation.

It goes to work.

It responds to traffic.

It makes purchases.

It reacts to weather.

It responds to economic changes.

It interacts with other agents.

And the city changes as a consequence.

This is the transition from:

Synthetic Data

to

Synthetic Society.


The Counterfactual City Engine

This is where the architecture becomes truly powerful.

Instead of asking:

What is happening in Dubai?

we can ask:

What would happen if we changed X?

For example:

Scenario A

Fuel prices increase 20%.

The simulator could estimate:

Car usage              ↓
Metro demand           ↑
Ride sharing           ↑
Traffic distribution   ↔
Household expenditure  ↑
Retail accessibility   ↓
Energy consumption     ↓
Enter fullscreen mode Exit fullscreen mode

Scenario B

Public transportation becomes 30% cheaper.

Metro adoption         ↑
Private vehicle use    ↓
Congestion             ↓
Energy demand          ↓
Station density        ↑
Commercial activity    ↔
Enter fullscreen mode Exit fullscreen mode

Scenario C

100,000 new residents enter the city.

Now the system can explore:

  • Housing demand
  • Traffic
  • Energy demand
  • Water consumption
  • School demand
  • Healthcare demand
  • Employment
  • Neighborhood pressure

before the policy is deployed.

This is the idea of a:

Counterfactual City Engine

A city government could test policies inside the simulation before testing them on millions of real people.


The City Brain

Once the Population Twin and Physical Twin are connected, we can introduce the next layer:

City Brain

The City Brain continuously observes the digital city and asks:

What is happening?

Why is it happening?

What happens next?

What happens if we intervene?

Which intervention produces the best outcome?

This turns the Digital Twin from a visualization system into a decision-making system.

Research published in 2026 is increasingly describing this transition: AI-enhanced digital twins are moving toward predictive intervention, foundation models, generative scenario creation, and autonomous management. (arXiv)


City Immune System

The same architecture can support another layer:

City Immune System

The city continuously searches for anomalies.

Normal State
     ↓
Continuous Observation
     ↓
Anomaly Detection
     ↓
Causal Diagnosis
     ↓
Risk Prediction
     ↓
Intervention
Enter fullscreen mode Exit fullscreen mode

Examples:

  • Unusual water consumption
  • Abnormal traffic
  • Sudden energy demand
  • Infrastructure degradation
  • Pollution spikes
  • Population displacement
  • Emergency congestion

Instead of waiting for failure:

Predict → Prevent → Adapt

The city becomes increasingly resilient.


Climate-Adaptive Cities

Now connect the Population Twin to the climate layer.

Suppose the system predicts:

Temperature will reach an extreme level tomorrow.

The City Brain can simulate:

  • Where people will concentrate
  • Which buildings will consume more energy
  • Where cooling demand will rise
  • Which transportation routes will become less attractive
  • Where public spaces become unsafe

Then it can optimize:

Cooling
Shading
Public Transport
Energy Storage
Water Allocation
Emergency Services
Enter fullscreen mode Exit fullscreen mode

This creates a city that doesn't simply measure climate.

It adapts to climate.


Why Synthetic Population Matters

There is a practical problem with building this system using only real-world personal data.

Large-scale individual-level data is:

  • Sensitive
  • Fragmented
  • Expensive
  • Restricted
  • Often incomplete

Synthetic populations provide a way to experiment without directly exposing real individuals.

This is increasingly important in urban AI research. A 2026 study on synthetic populations highlights the challenge of obtaining individual spatiotemporal activity data because of scarcity, skewed distributions, and privacy constraints. (ScienceDirect)

The goal is therefore not:

Replace reality with synthetic data.

The goal is:

Build a safe computational laboratory for urban experimentation.


From Smart City to Simulated City

This leads to a different definition of a Smart City.

A traditional Smart City asks:

How can we make city infrastructure smarter?

An AI-native city asks:

How can we build a computational model of the city that allows us to understand, simulate, and improve its future?

That is a much bigger ambition.

It connects:

IoT
+
Big Data
+
Synthetic Populations
+
Causal Graphs
+
Digital Twins
+
Generative AI
+
Agentic AI
+
Simulation
+
Optimization
Enter fullscreen mode Exit fullscreen mode

into one system.


The Long-Term Architecture

The final architecture could look like this:

                    ┌─────────────────────┐
                    │      REAL CITY      │
                    └──────────┬──────────┘
                               │
                         Sensors / APIs
                               │
                               ↓
                    ┌─────────────────────┐
                    │   CITY DATA LAYER   │
                    └──────────┬──────────┘
                               ↓
              ┌────────────────────────────────┐
              │       CITY DIGITAL TWIN        │
              │                                │
              │ Physical Twin                  │
              │ Population Twin                │
              │ Behavioral Twin                │
              │ Environmental Twin             │
              └────────────────┬───────────────┘
                               ↓
                    ┌─────────────────────┐
                    │   CAUSAL CITY GRAPH │
                    └──────────┬──────────┘
                               ↓
                    ┌─────────────────────┐
                    │   CITY SIMULATOR    │
                    └──────────┬──────────┘
                               ↓
                    ┌─────────────────────┐
                    │      CITY BRAIN     │
                    └──────────┬──────────┘
                               ↓
                    ┌─────────────────────┐
                    │ POLICY / DECISION   │
                    └──────────┬──────────┘
                               ↓
                         REAL CITY
                               │
                               └─────────────↺
Enter fullscreen mode Exit fullscreen mode

This is a closed-loop urban intelligence system.


Dubai as the First Laboratory

Dubai is particularly interesting for this approach.

Its highly diverse population, large expatriate component, multiple visa categories, industries, neighborhoods, and rapid urban transformation make it a challenging environment for population simulation.

The current Dubai Persona DAG explicitly models this complexity across 48 dimensions and generates a million synthetic residents calibrated against official UAE datasets. (GitHub)

That makes it a natural starting point.

But the architecture is not limited to Dubai.

The same framework could eventually produce:

Dubai Population Twin
       ↓
Abu Dhabi
       ↓
Doha
       ↓
Riyadh
       ↓
Singapore
       ↓
London
       ↓
New York
       ↓
Global City Simulation Network
Enter fullscreen mode Exit fullscreen mode

The ultimate objective would be a portable urban intelligence architecture, not a single-city model.


The Real Opportunity

The biggest opportunity is not selling another Smart City dashboard.

It is building the simulation layer underneath the city.

Imagine a future where a mayor, urban planner, energy operator, or transportation authority can ask:

“What happens if we introduce this policy?”

The City Brain generates thousands or millions of possible futures.

It evaluates:

  • Economic impact
  • Mobility impact
  • Environmental impact
  • Infrastructure impact
  • Population impact
  • Equity
  • Risk
  • Resilience

Then it recommends interventions.

The city becomes a place where policies can be tested before they become reality.


Dubai Persona DAG Is Step One

This is why I see the current repository differently.

It is not the final product.

It is a seed.

Dubai Persona DAG

→ Synthetic Population

→ Population Twin

→ Behavioral Agents

→ City Causal Graph

→ Counterfactual Simulation

→ Cognitive Digital Twin

→ City Brain

→ Living City OS

That is the trajectory.


The Bigger Idea

The future of cities may not be defined by how many sensors they have.

It may be defined by how accurately they can understand their own dynamics.

A truly AI-native city should be able to:

Perceive.

Remember.

Reason.

Simulate.

Predict.

Adapt.

Learn.

And eventually:

Heal itself.

The ambition is not to build a city with more technology.

It is to build a city with a computational nervous system.


From Smart City to Living City

The first generation of smart cities digitized infrastructure.

The next generation will digitize urban intelligence.

And the bridge between the two may be something deceptively simple:

a synthetic human.

One persona becomes an agent.

Millions of agents become a population.

The population interacts with infrastructure.

Infrastructure changes behavior.

Behavior changes the city.

The city generates new data.

The model learns.

And the loop continues.

That is when a Digital Twin stops being a mirror.

It becomes a world model.

And when that world model can reason about millions of possible futures before acting in the physical world, we are no longer talking about a Smart City.

We are building a Living City.


Open Source Starting Point

The first building block is available here:

Dubai Persona DAG — GitHub

The next question is no longer:

Can we model a city?

It is:

Can we build a city that can model itself?

AI #SmartCities #DigitalTwin #GenerativeAI #AgenticAI #UrbanTech #SyntheticData #UrbanPlanning #CausalAI #Simulation #CityOS #Dubai #ArtificialIntelligence #FutureOfCities

created by Seyed Alireza Alhosseini Almodarresieh

Top comments (0)