From Pen to Processor: Training AI to Translate Handwritten Housing Applications: A FormFest 2025 Profile

In the age of emerging artificial intelligence (AI), it’s easy to view tech upgrades as a mandate to scrap old systems. But it’s often smarter, and easier, to use technology to improve what’s already in place. At the Bay Area Housing Finance Authority (BAHFA), staff members Barry Roeder, Barbara “Babs” Deffenderfer, Glenn Brown, and Izzie Hirschy-Reyes trained AI to tackle the most analog problem of them all: deciphering other people’s handwriting.

The vast majority of applications in the nine-county region are filed online, mostly via mobile phone. This is a key feature of the Doorway Housing Portal, a regional clearinghouse of affordable housing listings that launched in 2023. However, around 5% are still filled out by hand, then mailed to a central P.O. box for processing. At first, those applications were entered manually, but the workload—and the tedium—quickly became too much to handle, giving Brown an idea.

“I said, you know, there are tools out there now that will take a PDF document and pull the text out of them, and they’ve gotten fairly good,” he said. “And then I began experimenting.”

The resulting process isn’t entirely autonomous—the P.O. box abides—but it is much faster. This is because it relies on open-source technology and commercially available tools, making it both cost-effective and easily replicable. 

“Other organizations use Microsoft—a lot of governments use Microsoft—or other platforms that have tools similar to [Microsoft] Power Apps, [but] what we’ve built is low-cost and a massive time-saver,” said Deffenderfer. 

Babs Deffenderfer: The Human Connection

Deffenderfer has worked on affordable housing for the past decade, but her connection to it stretches back to her teenage years, when she and her mom moved from couch to couch without a home of their own.

“That experience made me realize that anyone can fall into homelessness, and that I deserve a home as much as my homeless neighbor deserves a home,” she said. “That’s why I’m in local government, working to expand affordable housing opportunities.”

Today, Deffenderfer serves as BAHFA’s Doorway listings and partnerships project manager. Whenever someone chooses to submit a paper application—approximately one in 20 applicants do—it lands in a P.O. box around the corner from Deffenderfer’s office. After unlocking the mailbox door (“It’s not always full, but it’s rare that it’s empty”), she collects the mail, date-stamps and initials every application, then scans them into SharePoint as separate PDFs using a Canon copier. The files are automatically digitized into Power Apps forms, which she reviews for accuracy before forwarding to the housing portal.

Although this is a large workload, it’s easier than it used to be. Before AI, Deffenderfer had to manually type each application into the system, one field at a time.

“In my work, the digitization tool has cut the processing time in half,” Deffenderfer said. “I think we can further innovate by working to cut out some of the steps. However, at least for something as critical as a housing application, there will always be a need for human eyes to review the final version.”

Deffenderfer is happy to be those eyes. In many ways, it’s precisely why she came to government in the first place.

“I come from a long line of public servants. I grew up believing that smart, ambitious people who care about their neighbors work in government,” she said. “I can’t imagine doing anything else.”

Barry Roeder: Deciding to Care

The good news, according to Barry Roeder, is that there are a lot of housing programs. For the Doorway team, however, this also becomes the bad news—when there are different programs, there are also different forms.

“Within Doorway, we use a common housing application,” Roeder said. “But some housing providers don’t. We try to be flexible, and we don’t have the ability to make people do what we want region-wide, but we work on getting them there, because it works better for our partners and for applicants,[as] then they’re already familiar with the format.”

It’s a familiar situation for Roeder, who co-managed the development of San Francisco’s DAHLIA housing portal back in 2016. Even then, it was clear to him that a regional system would work better (“Housing is a regional issue!”), but it was equally apparent that local jurisdictions were not ready to pursue it. Instead, he focused on promoting the use of open-source code that would make it easier to integrate disparate systems into a larger portal down the road. 

Seeing both sides of an issue is a common theme for Roeder, who entered public service after his own experiences witnessing the Bay Area’s rampant income disparities. 

“I lived in a ZIP code where certain blocks had a $15,000 per-capita income, and in other parts, you’d see six-figure cars driving by,” he said. “I thought, ‘I can’t not see this anymore.’ The last project I did in the private sector around that time was to develop a line of credit for high net-worth individuals, and I was suddenly like, ‘I don’t think I care about high net-worth individuals.’”

He started by volunteering, then won a fellowship with the San Francisco Mayor’s Office of Innovation. While there, he heard that the city was looking for help developing its housing portal. 

“That set me on the path,” he said. “I just thought, ‘Somebody’s got to care.’ So, I decided to care.”

Isabelle Hirschy-Reyes: A Constant Direction

Some kids like LEGOs. Others like Pokémon. Izzie Hirschy-Reyes had different interests, and largely enjoyed “being in charge.” 

“I was always looking to direct things,” she explained. “I would direct my sister in plays. When I was 10 or 11, I created a whole business pitch to convince my parents to get us a [Nintendo] Wii. It had a finance proposal. I have always had very strong project management instincts, so it is extremely unsurprising that I’m in a role that does a lot of that.”

The focus on housing isn’t that shocking, either. Hirschy-Reyes started down that path at 14, when she began volunteering for a home repair nonprofit and saw firsthand the lengths people would go to in order to remain in their homes.

“The first time I went to volunteer, we were rebuilding a roof for a family living in a mobile home,” she said. “They had put tires on the roof to keep the tin on because it was falling apart—and it was such a shock for me. It just really compelled me to keep going, and to push myself to do more systemic interventions.”

As a senior product manager for Exygy, the outside vendor for Doorway, Hirschy-Reyes works on a variety of projects, from enhancing the platform’s technical infrastructure to redesigning the paper housing application. It’s a tech-heavy job, but Hirschy-Reyes is quick to note that the most meaningful improvements sometimes emerge from the least flashy projects.

“The universal form—that, in itself, is very innovative. It’s brand new to the region. We’ve never had a regional universal application for deed-restricted rental units ever before this,” she said. “All the tech pieces are really cool and make our life easy, but having one universal form that anyone can pick up, fill out, and submit—that makes a huge difference, and it didn’t exist before.”

Glenn Brown: Social Ethics and Computer Problems

As Doorway’s technical operations manager—and the guy who literally created BAHFA’s AI-powered application reader—Glenn Brown is, at least by trade, a tech guy. He started down that path the way most people do, while pursuing a PhD in social ethics.

“I started working part-time for a help desk when I was in grad school,” he explained. “Literally, it was me in a chair with a phone, trying to solve people’s computer problems, and I ended up being good at it.”

He initially used those problem-solving skills in the private sector, working for tech and finance companies. The pivot to public service came once he relocated from Chicago to the Bay Area and immediately realized the severity of the housing problem. Working on the Doorway portal and the application reader project allowed him to merge the two, leveraging this technology know-how to make housing more accessible. In its current iteration, the digitization system boasts a 94% accuracy rate, which means human oversight is still a must. But 94% is still pretty darn good.

“Every time you write a name, it’s pretty good at recognizing the name—better than I am, frequently,” Brown said. “That’s pretty incredible when you think about it. Things like that are important.”