ReadNOW

This article is part of “Build IT: Connectivity,” a series about tech powering better business.

Viome is aiming to transform disease detection, starting with the gut.

The Washington-based biotech startup offers at-home testing kits that analyze saliva, stool, and blood samples. Using RNA analysis, scientists at Viome can evaluate how genes and gut microbes are behaving in real time.

Once the tests are done, AI is applied to the results to generate personalized food and supplement recommendations. Users might be told to avoid spinach to reduce inflammation or take specific probiotics to support digestion and immunity.

So far, the company said it has sold more than half a million testing kits. Backed by Salesforce CEO Marc Benioff and venture capital firm Khosla Ventures, Viome is now scaling its tools to detect early signs of disease, including oral and throat cancer.

As Viome expands, the stakes are high. Grand View Research found that the global home care testing market is projected to grow more than 9% annually through 2030. As more consumers turn to medical testing kits for early disease detection and preventive care, the risks of misdiagnosis or ineffective treatment may surge if the tools aren’t built with precision.

To ensure its technology is both scientifically accurate and commercially viable, Viome relies on tight, ongoing collaboration between its research, engineering, and product teams.

In a roundtable interview, Business Insider spoke with Momo Vuyisich, Viome’s chief science officer, and Guru Banavar, the company’s chief technology officer, to discuss how the science and technology teams work together to deliver products that are ready for market.

The following has been edited for length and clarity.

Business Insider: Viome offers a range of products, including microbiome kits and early-stage cancer detectors. How do your science and tech teams work together to keep the AI models accurate, safe, and compliant?

Momo Vuyisich: It’s not just collaboration between science and tech — it’s a companywide effort. On the science side, we focus on three areas: lab work, data analysis, and clinical research.

Whenever we’re working on a health product, we rely on clinical research to guide development. This includes observational studies, where we learn from large groups of people, and interventional trials, where we test whether a tool works in real-world settings. For diagnostics, that means formal device trials.

In the lab, we use a method called metatranscriptomics, measuring RNA to understand what’s happening in the body right now. Unlike DNA, which stays the same, RNA changes based on things like diet or environmental exposure. That allows us to detect early signs of disease like inflammation or even cancer, based on how genes are being expressed.

We measure gene activity across human cells, bacteria, and fungi, and we also identify the types of microbes present in a sample.

Guru Banavar: What makes our approach powerful is the scale and detail of the data we collect. Each customer sends us stool, blood, and saliva samples, which we use to generate tens of millions of data points showing what’s happening in their gut, blood, and mouth.

Once that data hits Viome’s cloud platform, my team steps in. We use AI to figure out not just what organisms are present, but what they’re doing, like whether they’re producing anti-inflammatory compounds or if certain biological systems are out of balance.

We work with molecular data, which is far more complex than the text data most AI tools are trained on. So we use a range of machine learning methods, such as generative AI and algorithms that learn from labeled examples and draw insights based on patterns, where it’s appropriate. The key is using the right tool for the right problem, whether we’re detecting disease, recommending foods, or flagging health risks.

And because this work spans many fields, our team includes experts in biology, computing, cloud engineering, and more. Today, everything runs in the cloud, which allows us to operate at scale.

At-home medical testing and preventive health are fast-moving industries. How do you make sure you’re not moving too fast and overpromising on scientific outcomes?

Vuyisich: From the very beginning, we made clinical research a core part of how we operate. We didn’t just start building products. We started by measuring biological markers that were already published to impact human health, especially those linked to micronutrients. That was our foundation.

One of our earliest major studies was on glycemic response, how people’s blood sugar changes after eating. We spent millions of dollars running large-scale studies in the US and Japan, and we used that data to build machine learning models that predicted how a person would respond to certain foods. Afterward, we validated those models before we integrated them into our app.

We’ve followed that same process for everything from food and nutrition recommendations to our diagnostic test for cancer. We learn from both customer data and formal research, but the bottom line is we validate before we implement.

Banavar: On the tech side, we’ve built systems that help us move quickly while still being careful. We’ve automated a lot of the heavy lifting — like processing biological data and generating recommendations — so we’re not starting from scratch every time. When a new cohort of users joins Viome, we often retrain our models to reflect new biological data and ensure relevance. Some parts of that process are automated, but the final checks and tuning are still done by hand to make sure the model meets our standards before it goes live.

Another important piece is user education. Our app is designed to let people engage however they want, whether they’re just looking for simple guidance or want to dive deep into science. It’s an important part of making sure our customer base understands and can follow our recommendations.

Have you ever had to resolve conflicts between business priorities and scientific standards?

Banavar: Yes, and it’s natural in a multidisciplinary environment. We all come from different backgrounds. Biologists and machine learning engineers often describe the same process in totally different ways. Momo comes from the molecular side, I come from the computational side. Sometimes we talk past each other, meaning we miss things we say to one another that go beyond our domains of expertise. That’s why ongoing communication is so important.

There’s also the tension between speed and robustness. For example, when we’re building a new feature in the app, I’m OK launching a minimum viable product, MVP for short, which is a working prototype with basic functionality. But when it comes to health models, we won’t release them until we’ve validated the science. If it takes two more weeks to fine-tune, so be it. We’ll put a message in the app saying that a specific score, or a health indicator based on a user’s test results, is still being worked on — and that’s fine with me.

Vuyisich: It all comes down to defining what the MVP is. If it provides enough value for someone to pay for it and feel good about it, that’s the threshold. But an MVP for a toy can be rough and basic. An MVP for a cancer diagnostic needs to be very mature.

We don’t have a dynamic where business tells science what to do. We sit at the same table and make decisions together. If the science can’t hit the original target, we reassess. Can we lower the bar slightly and still provide value? If the answer is yes, we’ll launch.

The worst-case scenario is launching something that isn’t ready, but even that teaches you something. If no one buys it, you’ve learned a lot. Sometimes your friends and family say it’s amazing, but no one pays for it. That’s a signal.

But an even worse scenario is waiting too long for perfection. That’s buried more companies than anything else. If Apple had waited until the iPhone had all the features of iPhone 16, it would’ve gone out of business. Instead, they launched the first iPhone. They could be embarrassed today about how poor it was. But it worked. People paid for it. That’s what matters: bring it to market.

What lessons have you learned from building and scaling Viome that could help other companies trying to bring AI health products to market responsibly?

Banavar: First, there is no substitute for generating robust scientific data to support the value of health products. Second, when applying AI to health products, focus on areas and methods that can be independently validated and, ideally, interpretable, where companies can explain how the AI models reached their results to scientists, clinicians, and users. Finally, it’s possible, even in the health domain, to build products with an MVP mindset and implement a process for continuous improvement.

Vuyisich: Deeply understand the problem you’re trying to solve and identify a robust solution. At Viome, we set out to find the root causes of chronic diseases and cancer, which required measuring tens of thousands of human biomarkers relevant to health.

Also, use a method that’s accurate, affordable, and scalable. We spent over six years optimizing one lab test — metatranscriptomics — to go beyond the gold standard. This one test gives us thousands of biomarkers across multiple sample types with high accuracy.

Finally, it’s all about the people. Build a leadership team that deeply understands business and science, is aligned with the mission, and puts the company ahead of personal interests. Hire motivated, self-managed employees, train them well, and continuously coach them.