SCOPE
45 neighbourhoods · 9 villages · 9 synthetic personas · 3 modules
WORKING PROTOTYPE · CITIES · AI
Kentgörü
AI-powered, district-level policy decision-support system for local governments
A / VISIBILITY
45 neighbourhoods, one map.
Neighbourhood Focus shows the demo district (Bornova) as a 3D terrain map with an isometric bar chart beside it. When a layer is selected (population, income per person, household income, household size, marriage rate, education, share of women, business density, pharmacy access, socio-economic group, 2023 and 2024 election results) the map and the bars are recoloured together, and the ranking list reorders itself with the same data.
When a decision-maker clicks a neighbourhood, the map, the bar chart and the neighbourhood profile on the right update together. Which neighbourhood sits at which value reads at a glance, without searching.
B / NEIGHBOURHOOD PROFILE
One neighbourhood, one record.
For the selected neighbourhood, the right panel shows where it stands in the chosen layer (rank, difference from the district median, position between the lowest and highest values), then the gender split, income and household indicators, the education and marital-status distribution, and access to services such as pharmacies.
Below the panel sit a short synthesis specific to the neighbourhood and a "Run a scenario for this neighbourhood" button; the same neighbourhood context carries straight into the Scenario Lab.
C / PERSONAS
9 synthetic citizens, each can talk.
Neighbourhood data is read as nine synthetic personas formed by the intersection of income group (A-B, C1-C2, D-E) and demographic (young person, adult woman, adult man): Defne, Selin, Murat, Emre, Ayşe, Hasan, Elif, Fatma and Mehmet. Each persona's estimated population share in the selected neighbourhood is shown alongside.
Opening a persona reveals the profile (identity, description, media habits, decision criteria, a seven-trait personality scale), a knowledge bank and a one-to-one chat. Replies come with an inner voice, a positive/undecided/negative tag, an interest level and a theme; replies can optionally be given in the context of the selected neighbourhood.
D / SCENARIO LAB
Before the decision: 100 questions, 9 personas.
A strategy, service, activity or communication strategy is written; the scope (the whole municipality, villages, or a single neighbourhood) and the citizen profiles are chosen. The system proposes 100 questions across ten themes (cost and fees, transport and access, hours and timing, digital channel, trust and transparency, participation and voice, care and family, safety, content and benefit, public space).
The user approves, removes or adds questions. The approved questions are put to the nine personas; the answers are gathered into six output cards (expectations, likely reactions, neighbourhood-level assessment, access and inclusion, alternative scenarios, validation questions), a reaction matrix and a synthesis text. The result can be sent to the map, neighbourhood by neighbourhood.
100QUESTIONS · 10 THEMES
9SYNTHETIC PERSONAS
900ANSWERS / RUN
E / PUBLIC-OPINION SURVEY
From topic to result, in five steps.
The user gives a topic (for example "24-hour libraries in every neighbourhood"); the system proposes three survey drafts to suit it: a short pulse (5 questions), a detailed survey (10 questions) and an options comparison (6 questions). Questions are edited, reordered and approved; then the audience (the whole district, villages, or one neighbourhood), an income filter and the sample size are chosen.
The results screen gives each question's distribution with its margin of error, a breakdown table (income group, age, gender, education, urban/rural, 2024 vote, neighbourhood) and a CSV export. Every result is clearly marked "SYNTHETIC" on screen.
F / SCREENS
Five cuts through the interface.
01Neighbourhood Focus
The 3D terrain map, rotating bar chart, layer chips and the selected neighbourhood's profile on one screen.
02District overview
District indicators and neighbourhood comparisons: income against density, education against income.
03Question pre-approval
The 100 proposed questions are grouped into ten themes; each one is approved, removed or supplemented individually.
04Analysis flow
The process from data base to question generator, nine personas and analysis engine is followed live.
05Survey drafts
Three drafts for the topic: short pulse, detailed survey and options comparison.
G / WORKS ON A PHONE TOO
The same interface, in one column.
At 390 px the interface collapses to a single column: the map, layers, scenario and survey results work the same way on a phone.
H / ACCURACY AND LIMITS
Not real opinion, a preparation tool.
The personas and survey respondents in Kentgörü are synthetic profiles weighted by the neighbourhood data of the demo district, Bornova (TÜİK 2025, Endeksa, YSK). The outputs are not real citizen opinion or field data; they exist to show which questions should be put to which groups when a new policy is considered.
That is why results are marked "SYNTHETIC" on screen and why, on every run, the system generates validation questions that a real study should settle. Kentgörü is a design prototype; moving it to a production architecture is a separate decision.
I / SCOPE AND TECHNOLOGY
From the map to the decision card.
DATA
TÜİK 2025 · Endeksa · YSK (2023 general, 2024 local elections)
MAP AND 3D
MapLibre GL (terrain map) · three.js (persona avatars) · SVG bar chart
RUN MODE
Fast in-browser simulation · generation with Claude (in the development environment)
OUTPUT
Six output cards · reaction matrix · synthesis · CSV · export to map
INTERFACE
Turkish · desktop and phone · HTML, CSS, JavaScript