San Francisco’s landscape of ornate, often colorful, Victorian homes is famously beautiful.
But which one is the most eye-catching?
That’s the question Gen Z tech founder Sarv Kulpati set out to answer on a recent weekend night by coding a website that crowdsources the city’s most attractive buildings, asking users to rank one over another.
“It’s quite literally ‘Hot or Not’ for S.F. buildings,” he said.
Kulpati is calling his website, which has garnered more than 16,000 votes since he posted it on the social media platform X last week, “Facade Mash.”
The name is a play on “Face Mash,” a similar — though probably more controversial — project that Mark Zuckerberg designed, enabling his Harvard classmates to rank each other based on looks. The concept is somewhat comparable to a complex ploy Larry David might conjure up in Curb Your Enthusiasm, Kulpati said.
“It was almost a tongue-in-cheek kind of thing with my friends,” he told KQED. After moving to San Francisco a year ago, the 24-year-old was inspired to sketch more of the beautiful architecture he passes while biking through his neighborhood near Haight Street.
He considered going for a walk or a ride for inspiration, but thought better of it.
“I don’t want to draw a seven out of 10 building. I want the best,” Kulpati said.
Instead, he scraped images of 10,000 buildings from Google Maps, used AI to ensure each was only displayed in a single front-facing image, and coded a website before blasting out a link on X — “which, arguably, is even more effort,” Kulpati said.
The site is one of the latest in a growing wave of tech experiments designed by young San Francisco programmers, many of whom are using AI to help create gratuitous online projects that can spread quickly across social media.
Last year, Kulpati was among a group of more than a dozen friends who designed a monthlong, citywide scavenger hunt called PURSUIT. Another one of that game’s creators, Riley Walz, has pulled off a number of other stunts, including an app that tracked San Francisco Municipal Transit Agency workers ticketing parked cars in real time and another that chronicled the music passersby listened to on an undisclosed street in the Mission District.
Kulpati said the growth of these projects reflects a trend in the tech industry.
“I’d say most programmers now spend most of their time prompting AI to help them write code,” he told KQED. “One way to look at that is, you can make your same old boring stuff faster. Another way to look at it is: ‘What is stuff you’ve never made before that now you can make?’”
Designing Facade Mash would have been a weeklong endeavor, at least, if Kulpati had to code it himself. Instead, he said, he built most of it in one night.
“There’s definitely this pocket of, I’d say, creative technologists who are using this stuff and applying it in interesting ways,” he said. “It’s very S.F.”
Kulpati described the process of ranking buildings as a kind of collaboration between humans and AI. He used the computer to do an initial order before inviting people to play “this or that,” so they wouldn’t be comparing an empty lot, per se, to a Sea Cliff waterfront home.
Still, it’s been evident that computer systems don’t have the same taste as the human eye.
“Civic Center was the best,” after the AI ranking round, he told KQED. “If you look at the leaderboard now, it’s more cozy.”
The website asks users a simple question: Which building looks better? Below the prompt are two photos of addresses somewhere in the city. Pick one, and two fresh facades appear.
Right now, the public leaderboard is dominated by classic Victorian homes, many of which are located in Lower Pacific Heights, Western Addition and Alamo Square, according to the map of hot spots also available on the site. But there are some outliers: as of Friday, No. 17 is a boxy commercial building on Van Ness Avenue painted with an abstract mural, and a small one-story building tucked on a side street off of Patricia’s Green in Hayes Valley, covered with a collage of colorful knickknacks and art ranks 43rd.
Kulpati said the project has mostly succeeded, pulling out gorgeous homes that represent a fairly unified taste among voters. But he’s still on the hunt for more of the “really weird” hidden throughout the city.
He thinks that might emerge if he’s able to do a comprehensive sweep.
Right now, the project is capped at the maximum number of buildings Google Maps would let him scrape for free, but he said he’s thinking about expanding the scope.
“I’m too prone to side projecting,” Kulpati said. “This is almost like I shouldn’t, but maybe I will.”
To learn more about how we use your information, please read our privacy policy.