How a 10-Person Startup Built Enterprise-Grade Biometric Security

Three engineers. Ten people total. Ten patents on proprietary biometric matching and liveness detection. A palm dataset on par with Amazon’s. A live deployment on a crypto exchange serving 32 million users.

Those numbers don’t look like they belong to the same company. Zach Meltzer, CEO of VeryAI — which builds palm biometric identity verification from smartphone cameras — has an explanation for how they fit together: the old headcount math doesn’t apply anymore.

The Old Equation Is Broken

Three years ago, a company building enterprise biometric security with 10 patents would need at least 20 to 30 people. Dedicated front-end engineers, back-end specialists, mobile developers, security researchers, QA teams. Each function siloed, each specialist expensive.

Meltzer estimates his team would have been “at least double in size, maybe even triple” if he’d built it a few years earlier. The difference isn’t just AI coding assistants writing boilerplate. It’s a structural change in what one person can own.

“You can scale quickly with just a few engineers and build everything that you need at maybe a 5X speed,” he says. That multiplier isn’t theoretical — it’s reflected in the patent count, the dataset size, and the production deployment.

The Swiss Army Knife Model

VeryAI doesn’t hire specialists. They hire people who can work across the full stack and adapt to whatever the company needs that week.

“It’s really important for engineering teams to have holistic approaches now,” Meltzer explains. “Less focused on just back end or front end, and really have experienced engineers who can fill in the gaps across the board.” He calls it the “Swiss army knife” model — engineers who can shift between mobile development, model training, and infrastructure depending on what’s most urgent.

This isn’t just an engineering philosophy. Meltzer applies it across the whole company. “You really want people to be like a Swiss army knife in many senses so that they can use AI to complement their skillset rather than fully rely or double down on one piece of building the company.”

The hiring implication is significant. When you’re looking for generalists who can use AI tools to cover multiple domains, the interview process changes. VeryAI leans heavily on qualitative evaluation — can this person reason through unfamiliar problems, not just execute within their specialty?

“The way that you evaluate a candidate has changed to a lot more intangible skills,” Meltzer says. Take-home assignments, once a reliable signal, have become unreliable. “It’s very easy to complete a take-home assignment now and fully feed that through AI.”

The Data Moat as Force Multiplier

A small engineering team can ship fast, but can they build a defensible product? In VeryAI’s case, the moat isn’t in the model architecture — it’s in the data.

When asked what a developer would miss by prototyping palm biometrics from open-source models and public datasets, Meltzer’s answer is blunt: “Even if you have a small sample of Palm images, it’s going to be incredibly difficult to reach a level of accuracy that is actually sufficient to build a real product.”

Open-source palm recognition models exist. University datasets with real palm images are available. CVPR papers describe the underlying techniques. But the gap between a prototype and production is the data volume required for reliable matching at scale.

“We have the largest data set of Palms in the world on par with other companies. Only other companies on par would be like an Amazon,” Meltzer says. “Some of these other data sets that you might see are simply not enough to create an accurate enough solution for building a model. And that’s why no one else has really done it yet to scale.”

This creates an unusual dynamic: a tiny team with a massive data advantage. The three engineers aren’t competing on code — they’re standing on a dataset that would take a larger competitor years to accumulate. The patents add legal protection. Together, data and IP form the kind of moat that doesn’t require headcount to maintain.

What This Means for Startup Builders

The VeryAI example isn’t prescriptive — not every domain rewards the small-team-plus-data-moat structure. But it illustrates a pattern that’s becoming more common. AI tools reduce the minimum viable team size. Data and IP advantages become the defensible layer, not engineering headcount. And hiring for adaptability matters more than hiring for specialization.

Meltzer’s bet is that this trend accelerates. “These highly specialized roles might start to go away. Having people with very broad skill sets is going to be incredibly important to companies as we continue to leverage more AI and focus on building smaller, leaner teams.”

For founders still staffing like it’s 2022, the math might be worth revisiting.

FAQ

How many engineers does VeryAI have and what have they built?

VeryAI operates with a team of 10 people total, including 3 engineers. They’ve filed 10+ patents on palm biometric matching and liveness detection, built a dataset on par with Amazon’s, and deployed their SDK on MEXC — a crypto exchange with 32 million users.

What is the Swiss army knife hiring model for AI startups?

Instead of hiring specialists for front-end, back-end, and mobile separately, companies hire experienced generalists who can work across the full engineering stack. AI tools let each person cover more ground. Zach Meltzer says this applies to business roles too — not just engineering.

How has AI changed the way startups interview engineering candidates?

Take-home assignments are no longer reliable signals since candidates can run them through AI. VeryAI has shifted toward evaluating qualitative and intangible skills — reasoning ability, problem-solving approach, and adaptability — rather than just quantitative output from coding tests.

Can you build palm biometrics from open-source models and public datasets?

Open-source palm recognition models and university datasets exist, but they lack the data volume needed for production accuracy. VeryAI says their dataset is on par only with Amazon’s. The gap between a prototype and an enterprise-grade product is the data, not the model architecture.

How does VeryAI’s small team build enterprise-grade security products?

Three factors: AI-augmented development at roughly 5x the speed of traditional teams, proprietary data that forms the core competitive moat, and 10+ patents protecting their matching and liveness technology. The team doesn’t need to be large because the defensible advantage is in data and IP, not headcount.

What kind of data moat does VeryAI have in palm biometrics?

VeryAI claims to have one of the largest palm datasets in the world, comparable only to Amazon. They don’t use synthetic data — all training data comes from real palm scans. This dataset took years to build and creates a barrier that new competitors can’t easily replicate regardless of team size.

Are specialized engineering roles disappearing at AI startups?

VeryAI’s CEO believes highly specialized roles are declining in importance as AI tools let individuals cover broader scope. Companies are hiring people with broad skill sets who can use AI to complement their capabilities across multiple domains rather than focusing narrowly on one area.

How many people did a biometric security startup need before AI tools?

Meltzer estimates VeryAI’s team would have been “at least double in size, maybe even triple” — 20 to 30 people — just three to five years ago. AI development tools let the current team of 10 accomplish what previously required significantly more specialized headcount.

How does VeryAI find and recruit for a 10-person team?

Most hiring comes through personal relationships — the CEO’s network and the existing team’s connections. As the company grows, they expand to recruiting firms and headhunters. The emphasis is on finding people who’ve demonstrated adaptability across multiple domains rather than deep specialization in one.