Offline demo · Census/PUMS demographics · synthetic residents · fixed fixture outcomes, not answers to new questions. Exit demo
37° 46′ N
SIMULATED CIVIC FIELD
waking the city…
PREDICTING
Audience
ask
filter residents
Build a new synthetic sample using Census records that match every selected trait.
Vey
Know the impact of a million-dollar decision in twenty minutes instead of four months.
A live simulation of San Francisco: 10,000 digital residents built from real US Census data, each with a background, motivation, memory and daily life.
1Simulate
Ask the city anything.
Type a question in the box at the bottom, e.g. What if I increase the price of Chipotle bowls by 20%?
Residents answer individually; you get the split (keep buying · buy less · switch · stop) and the reasons.
A/B: compare two messages with the same audience.
2Filter
Ask exactly who matters.
Tap the audience pill: age, neighborhood, occupation, education, e.g. 25-year-old software engineers.
The result shows who was asked and how many Census records back it.
3Understand
See the people behind every number.
Under each result: bar, histogram, scatter, box or line by demographic.
Hover a bar to see the residents; click it to read each person’s own answer and why; click a person for their full profile.
4Remember
The city evolves as you use it.
Post news or events in the Timeline (A 5.9 earthquake hit the Bay Area); residents react in character.
Every event and survey is remembered and shapes the next answer. Ask the same question again and the timeline shows the shift.
Everything you post and ask lives in the Timeline; verified Census questions (show the age distribution) work too.
Track record
2024 presidency: 83.8% actual vs 81.3% predicted. March 2024 Prop A: 70.4% vs 70%.
Backtested with a model whose knowledge cutoff predates both events, so nothing was memorized.
Built by Nandan Pericherla, Abhay Sudhir, and Mahin Bharathwaj.