Kat Wifvat, PhD in applied math and founder and CEO of Nuubi, explains how AI teaching assistants are filling the gaps traditional help forums never could, and why student data privacy has to come first.
Almost 40% of college students in the U.S. never graduate, and a third of them drop out simply because they had academic questions they couldn’t get answered in time and started falling behind. That statistic is what pushed Kat Wifvat to leave the classroom and build a company around it. It’s not a technology problem in the way most people assume. It’s a support gap, and Kat thinks AI is finally positioned to close it without replacing the human side of teaching.
Kat Wifvat has a PhD in applied math and has taught as an adjunct data science professor at multiple universities. She’s also the founder and CEO of Nuubi, an AI-powered Q&A platform built for higher education. She joined us on Deconstructing Data to talk about where agentic AI tutors are headed, what they can actually do that a generic chatbot can’t, and why student data privacy has to be baked in from day one.
Why do students avoid the help forums that already exist?
Every online class basically has a Q&A forum built in, and Kat’s observation from years of teaching was that almost nobody uses them.
“Students are too shy to ask for help. The teacher doesn’t get notified until they get an email anyway. The students who are doing well, they don’t have any incentive to help. So it just turns into the system where nobody uses it.”
— Kat Wifvat, Founder and CEO, Nuubi
That’s the actual problem Nuubi was built to solve. It’s a group Q&A forum where an AI teaching assistant, what Kat calls the AITA, provides a first-pass answer to every question the moment it’s posted, at 2 a.m. or whenever a student is actually stuck. The professor gets notified immediately instead of finding out days later, and students who jump in to help each other earn points for leadership and communication.
“Post here, I will get notified of it immediately so I’m aware that you had the question. You get an immediate response that is better than some other chat tool, and then other students who jump in, they all actually get rewarded for leadership and communication points.”
— Kat Wifvat, Founder and CEO, Nuubi
That last part matters more than it sounds like. Recruiters consistently say leadership, communication, and teamwork are some of the hardest skills to find in candidates, and Kat built incentives directly into the platform to surface exactly that.
What can an agentic AI tutor actually do that generative AI can’t?
This is where the conversation got genuinely interesting. Kat drew a clear line between generative AI tools that just produce text, and agentic AI that can audit, fact-check, and take action.
“A lot of those generative tools, we’ve come up with this word for it, hallucinating. Those tools, they’re not 100% accurate all the time. They were not built to solve mathematical problems necessarily. So now the question is, how do we build on those systems to have a layer that’s auditing what’s going on, fact-checking, and then making decisions?”
— Kat Wifvat, Founder and CEO, Nuubi
In practice, that means if a student hasn’t checked in or asked a question and never got an answer, the system doesn’t just sit there. It can send an email, loop in a student support advisor, or flag the instructor with specifics on what topics students are actually struggling with. That’s a meaningfully different job than generating a paragraph of text.
Kat also flagged something professors deal with constantly but rarely get ahead of: students who are participating a lot but are quietly at risk of burnout. The instinct is to assume high engagement equals doing fine, but her data has shown that’s not always true.
“There’s some students that we found actually that are participating a lot but are at risk of burnout. And so we might not think they need support because they’re participating, but they actually also do need extra support.”
— Kat Wifvat, Founder and CEO, Nuubi
How do you personalize AI tutoring without compromising student privacy?
This is the part every school and every vendor building in this space has to get right, and Kat was direct about where the line sits. Universities do not want student data used to retrain models, full stop, even when identifying information has been stripped out.
“They really want the data to already have been trained somewhere else or on people who volunteer for that. They do want the models to not be retrained using any student data.”
— Kat Wifvat, Founder and CEO, Nuubi
Nuubi’s approach is to assign each student an anonymous ID that only the instructor can map back to a real name, purely for tracking who needs help. The AI can pull from a student’s stated interests, say they mentioned loving sports in an intro post, to tailor how it explains something, without that information ever being used to change the underlying model.
“It’s basically able to just generate text where it’s grabbing information, but it’s not saving any kind of new information or storing any kind of new information. Learning how to change its responses in the way it responds over time, that would involve some of that retraining, versus just taking data that’s already available, but the model itself isn’t changing.”
— Kat Wifvat, Founder and CEO, Nuubi
Kat expects the field to move toward an opt-in model similar to cookie consent banners: students choose whether their course materials and interests get used to personalize their AI experience, understanding that opting out just means less-tailored help, not none at all.
Is the traditional four-year degree still the right model?
We got into this because it’s the question hanging over every conversation about AI and education right now. Kat’s take wasn’t that college is dying, but that the path around it is fragmenting.
“Gone are the days of just you go to school, you get a four-year degree, and then you work the same job for decades. We are rapidly seeing a lot of curriculum changes that are really embracing AI and really helping students be prepared for an AI-ready workplace.”
— Kat Wifvat, Founder and CEO, Nuubi
Her prediction is a rise in micro-credentials and a resume that behaves more like a living document than a fixed record, alongside a lot more upskilling and reskilling happening outside the traditional four-year system entirely, even as peer groups and structured curriculum continue to prove genuinely valuable for the students who go through them.
The bigger picture
What stood out most in this conversation is how much Kat’s approach resists the obvious shortcut. It would be easy to build a tool that just answers homework questions faster. Instead, Nuubi is engineered around the actual reason students drop out: they get stuck, feel isolated, and don’t ask for help in time. The AI’s job isn’t to replace the professor or write the essay. It’s to catch the moment a student is falling behind before a whole week passes and the problem snowballs, while keeping their data locked down tight enough that trust in the system doesn’t erode along the way.
That’s the same principle that should guide any company building with AI and personal data right now: the technology only works if people trust what’s happening with their information behind the scenes.
To connect with Kat, find her on LinkedIn or check out her work at nuubi.com.
And if you’re thinking about how to responsibly use data to understand and support your own audience, that’s exactly what we do at BDEX. Visit bdex.com and click “Talk to an Expert” to get started.
Video
Watch the full episode: AI in the Classroom: Preparing Students for the Future of Work
This article was adapted from an episode of Deconstructing Data, BDEX’s weekly podcast on data-driven marketing. Tune in live every Thursday at 4:15 PM Eastern on LinkedIn.
About BDEX: For companies that need clean identity data to power their products, BDEX offers unmatched quality and execution. Visit bdex.com and click “Talk to an Expert” to get started.
