The Center for Data Innovation spoke with Max Versace, co-founder and CEO of Neurala, a Boston-based company that makes AI-powered software to help manufacturers improve their visual inspection process. Versace discussed the origins of the company and his earlier work in developing neural networks, the unique features of the technology that Neurala uses to make visual inspections more efficient, and the potential impact that rapid advancements in AI will have on Neurala’s technology.
This interview has been edited.
Martin Makaryan: What does Neurala do, and what pushed you to co-found the company?
Max Versace: Neurala makes software that uses AI to automate visual inspection processes in manufacturing facilities. We work with manufacturers in different industries, from automakers to those in the food production process, to train a deep learning model that powers our customized AI software. Then, we integrate the software into their existing hardware for machine vision, expediting detection of manufacturing defects on virtually any physical product. Our software also offers a feature to ensure that all components of particular interest in a product are in the right place and in the right orientation, without any defects. This feature can run from a single inspection camera, reducing the cost per inspection without causing any delays.
Neurala’s beginning dates to my student years when the company’s co-founders and I were studying neural networks. We founded Neurala in 2006, but AI was not as popular or well-known at that time. It has been an interesting and somewhat surprising journey to see the evolution of AI over the past two decades. I am surprised that it took us as a society such a long time to get to today’s more advanced AI systems, such as large language models (LLMs) and generative AI.
The initial focus of my research was on robotics and automation, but I quickly realized that due to the relatively short supply of robots, it would be better to specialize in a more niche area where applying deep learning algorithms could yield more tangible benefits for society. That is why we decided to focus on cameras. We had already rolled out smart consumer devices, such as drones and some robots, and our focus was to automate them. Eventually, we saw a need for software solutions that would allow users without expertise or experience in AI to become more productive in their work. That is how we evolved into a company that empowers factories to automate their visual inspection processes and access physical environments or products that are hard to examine with just a human eye, such as inside complex industrial machines.
Makaryan: How does Neurala’s technology ensure accurate detection of product deficiencies?
Versace: In contrast to traditional machine vision technology, which requires a developer to input a certain number of constant rules to program a machine for a certain function, we have worked hard to create a technology that largely emulates the process by which a brain learns to perceive, understand, and respond. This is the foundation of deep learning models, of course, but what sets Neurala apart is that our AI tool continuously updates and learns from new data as clients make changes to their products or introduce new ones, allowing for the most accurate detection of deficiencies. Once our clients deploy the software, it automatically updates to reflect the changes in the production line.
Makaryan: What is the role of data in Neurala’s business model and where does it come from?
Versace: There is no AI without data, and there is no Neurala without AI. Data is at the core of what we do, but the way we approach it may be slightly different than other AI companies, especially those developing LLMs or popular products like chatbots that the public probably is more aware of. Neurala develops custom technologies for each of our clients based on the unique features of their products and business needs. As such, we work with the datasets that our clients supply to us. This confidential data includes variables on key product features, and we pride ourselves on our integrity in keeping our clients’ data secure. Regarding the time it takes to develop a new AI system for a client, it can take anywhere between seconds to days in some cases to train the algorithm on their dataset. The timing also depends on the quality of a customer’s data. In some cases, we can deploy a custom AI system as quickly as a day.
Makaryan: How will rapid advancements in AI shape Neurala’s future course?
Versace: As someone who has engaged in researching AI and deep learning for decades before the current AI boom began to unfold, it is very satisfying to observe hundreds of new companies that use AI to benefit society emerge. What that means from a business perspective, however, is that it has become somewhat harder to convey to the customer base exactly what we are doing and how it sets us apart from other AI companies. Regarding Neurala’s future trajectory, I don’t believe that the kind of advancements we are seeing, especially in generative AI, will significantly change our technology. We have been using deep learning algorithms for manufacturing quality control for years now, before the hype around AI started taking over virtually every industry in recent years. Nonetheless, I believe there are some areas where improvements in current AI technology and potential new breakthroughs could enhance our software. Specifically, I see potential improvement for customers to use natural language to interact with AI technology, including our software, making it even easier and more appealing to use Neurala’s product.
Makaryan: What was the biggest challenge you encountered as a data innovator?
Versace: Transitioning from the academic world to the business world and commercializing innovation was the biggest challenge for me. As a professor, my main priority was innovation itself, which meant that I had to overcome the real-world challenges of turning an idea into an operational business. When you come from a technical background and think that you have created this fantastic new tool, without thinking holistically about the business side of the equation, I think that you will have a painful reckoning.
Another challenge as an innovator that I’ve encountered is the need to find my niche. Often, as an innovator or academic, we want to focus on the more “fun” things, such as robotics or drones, or work hard to create a technology that people can use universally. This intellectual curiosity and the desire for invention can hinder an innovator’s ability to identify a real, plausible business opportunity that can turn their idea into a tool to actually change people’s lives positively. This has been another major challenge and a choice that we have had to make at Neurala at various points in the company’s history.
