Those internet daydreams about buying a home in the country might take on a new element of realism, courtesy of an experimental AI-infused graphics technique.
Zillow is turning drone photography into immersive aerial views of homes on its platform through a combination of AI and a goofily named visual rendering method called “3D Gaussian splatting” that’s more commonly used in video game development.
“Imagine being able to spin around the exterior of the home, going up and down in elevation, going in and out in proximity to the home,” Zillow CTO David Beitel told Tech Brew. “The AI models take that drone footage and then build the internal structures and computer vision models to allow that to be realized within our app and our website.”
It’s the kind of visual machine learning capability that Zillow has built much of its AI chops around—that and its famous Zestimate market value calculator. Before ChatGPT made AI a permanent fixture in business headlines, Zillow had assembled a research team laser-focused on letting people virtually immerse themselves in far-flung properties, inside and out, through their phone or computer screens.
“I would say a high percentage of the patents that Zillow has are driven off our over 10 years now [in] research and development with computer vision, first starting with more of the interiors, floor plans, layouts and interior walkthroughs, as well as now the exteriors with SkyTour,” Beitel said.
Now, like virtually every other company, Zillow has its sights set on generative AI. The guiding mission is to move Zillow from a home listings directory to a platform that aids in every step of complicated real estate transactions, Beitel said.
“Our shift [is] from just helping people find homes to really thinking about that integrated transaction experience, helping them all the way through. We don’t consider it successful until you’re in that home—and joyfully in that home. And then how can AI really enhance that?” Beitel said.
The stack: Zillow has started exploring GenAI features like AI-powered FAQs for each property on its StreetEasy platform, personalized AI-generated intro messages between shoppers and agents on StreetEasy, and AI call summaries for agents to recollect previous interactions with a customer.
In addition to Zillow’s proprietary models, Beitel said the company works with most major model providers, as well as enterprise tools like Glean—for internal AI search—and Cursor and Claude Code for coding assistance. With so many options, it can be difficult to pick the right model for the right job, he said.
Keep up with the innovative tech transforming business
Tech Brew keeps business leaders up-to-date on the latest innovations, automation advances, policy shifts, and more, so they can make informed decisions about tech.
“It’s tricky to do this, because the models are changing so quickly,” Beitel said. “We have partnered very, very closely with the big players and have enterprise agreements with pretty much all of them, both for utilization internally as well as with our products.”
A research mindset: Zillow’s machine learning and AI research team details some of its thinking in long technical blog posts on its website. Beitel said Zillow is careful about what technical research it reveals, but allowing researchers to publish papers and attend conferences helps draw academic talent and raise the team’s profile.
“We have to be a little careful about…what are we keeping more private and what are we patenting,” Beitel said. “But we’re very proud of the stuff we’ve published. We just recently [went to a] computer vision conference and had a number of papers accepted at the highest levels. And our speakers are interacting with some of the top in the field.”
The guardrails: Building AI for anything housing-related can be fraught with risk. In addition to laws designed to prevent discrimination in sales, rentals, and financing, there are all kinds of trust and safety concerns when it comes to such a high-stakes application of AI.
Beitel said Zillow has an internal AI ethics team that ensures that “our algorithms are reviewed and audited very carefully.” Last year, the company rolled out an open-source tool called the Fair Housing Classifier designed to detect questions in searches or chatbots that might lead to discriminatory responses.
What’s next: Zillow is currently working on AI agentic skills—not to be confused with the human real estate agents the platform serves—but Beitel said he can’t share much about that yet.
Beitel said he doesn’t see AI agents putting their human counterparts out of business. “We fundamentally believe that humans are very important. Agents are very important to this process. I mean, you can imagine that first-time home buyer who’s really nervous,” he said. “The agent is really there to help shepherd the buyer and seller through this process, and so our vision is, ‘How do we think about leveraging technology to allow that buyer and seller to show up very empowered to that conversation?’”