University of North Carolina Chapel Hill students Krish Vazirani and Aryan Aladar met about a year before the time of this writing. Both had experience building software, and together they decided to explore ways in which AI could help address gaps they’d identified in traditional testing.
In short time, this shared interest led them to develop their own alternative, using AI personas to simulate user behavior, test product flows, and pinpoint issues. Altogether, this amounts to a faster and cheaper alternative to traditional user testing. The solution evolved into their own co-founded startup, dubbed Swarm.
“Currently, companies use their real users as lab rats. That’s the equivalent in science to using real humans as your test subjects without doing clinical trials,” Aladar said.
“So what we do is we give them AI personas which are like lab rats, which you can test as many times as you want, but [companies] still eventually go to the clinical trial with the humans. We give early validation before you have to do that and waste money.”
The Hive
At its core, Swarm functions as a testing layer. The platform generates AI personas—digital users built either from synthetic data or a company’s existing analytics—and deploys them through an application’s interface to replicate real-world behavior.
These personas are drawn from a database of roughly a million demographic profiles, which serve as templates. From there, Swarm selects clusters based on a company’s target user and feeds in psychographic inputs, positioning the personas to mirror how different users might navigate a product.
For example, a persona could mimic a DoorDash driver trying to use an application on their phone, with one hand, in bright sunlight. Using synthetic data, teams can generate dozens or even hundreds of personas like this at once, creating a large sample of simulated users without needing real customer data.
“That works really great for startups or early teams who don’t have an immense amount of data, so it’s great to just spin up synthetic data and use that as a baseline,” Vazirani said. “But with more established teams that have large data models in the background, we can pull that analytics data and cluster it into different segments.”
Indeed, for larger companies, Swarm uses preexisting data to segment users and simulate how different groups behave.
“Say, for example, Amazon has been collecting product analytics for years, and they give us some of that data and we start segmenting it into clusters,” Vazirani said. “And then we basically create digital minds or twins of your real users online at your disposal.”
Finding the breakdowns
Once deployed, these personas move through a product’s interface like real users, clicking through flows, entering information and navigating key features. As they do so, Swarm identifies what works and, more importantly, friction points and breakdowns—from missing features to areas where users are likely to drop off. It then returns feedback explaining what went wrong, why, and how it can be improved.
In a pilot with a Launch Chapel Hill company preparing a mobile app, the platform flagged multiple critical bottlenecks, including a missing “resend verification email” option in the onboarding flow, ultimately helping the company resolve issues before release.
Key differentiators
Unlike many existing testing tools, Swarm is built to operate across the entire product lifecycle. It can be used on early-stage mockups, live applications or local development environments, enabling teams to test continuously as they build.
Crucially, Swarm’s personas also retain memory as a product evolves, rather than operating independently within each test. That allows the same set of simulated users to track changes over time, evaluating not just what is broken, but how updates improve or introduce new issues across iterations.
“When you have a completely new persona, you lose a lot of resolution or important information from your past runs,” Aladar said. “But here you have the same personas looking at it again and again, they’re telling you what went wrong, what you fixed, and then what went wrong again.”
Momentum and the future
Swarm was built this year, and already the team has raised funding, secured pilot customers and begun working with enterprise-level companies, including KPMG, which has purchased licenses for use across its startup cohorts (Swarm itself recently participated in the Launch Powered by KPMG accelerator program.)
QUICK BITS
Startup: Swarm
Co-Founders: Krish Vazirani & Aryan Aladar
Founded: 2026
Team size: 2
Location: Chapel Hill
Website: useswarm.co
Funding: $100K+ in funding raised from LeapYear & Afore Capital
Over the next few months, the team is focused on adoption and revenue, with a goal of having around 100 users running tests daily. They aim to soon be integrated into development workflows so Swarm can run automatically, return feedback and, eventually, fix issues itself without manual input.
Longer term, the founders see the platform expanding beyond UI and UX into a broader validation layer—one that can test not just usability, but whether a product works, is secure and delivers value. At the highest level, they envision Swarm extending into other industries, becoming a tool that can simulate outcomes before decisions are made.
“I think in 10 years our engine will become so good that we can almost validate and simulate any decision out there,” Aladar said.
