Shikhar Shrestha has been building security systems since he was a teenager. It began as part obsession, part coping mechanism. He'd been traumatized when he and his mother were robbed at gunpoint when he was 12. The area of his hometown in eastern India seemed to have lots of security cameras--but what was the use? Help did not come while he was being threatened, and while his mother's jewelry was being stolen. He thought about that a lot.

As a child, Shrestha tinkered with technology, including building homemade security systems for neighbors. Years later, he enrolled at Stanford, doing graduate work in electrical and mechanical engineering. There he met computer science grad student Vikesh Khanna--and the pair had a light bulb moment in conceptualizing the future of video innovation.

"We had an idea that artificial intelligence and video technology were getting so good that in five years video tech and A.I. could look at a video more exactly than humans can," Shrestha, now 30, says. "If any camera out there can tell you right away when it sees something suspicious, that would make for a great security system."

The pair earned master's degrees, and in 2017 founded, iterating on their idea with funding and support from the Silicon Valley startup incubator Y Combinator. They had a clear goal: to prevent every physical security incident possible. They developed a technology that combines A.I. and a computer-vision breakthrough, called computer vision intelligence, to understand situational context. It could, in real time, identify elements in a video from a human walking, to a car tailing another car, to a weapon being brandished, to a perimeter breach.

The founders thought they had a straightforward problem to fix. With conventional enterprise security systems, video cameras capture an endless stream of video--which is rarely, if ever, watched in real time to actually stop, prevent, or quickly respond to an incident. During his time in Y Combinator, Shrestha sent 100 emails a week to security chiefs at large companies, hospitals, hotels, and governments, to learn more about his market and its needs. He quickly learned that no one wanted a new security system--they already had cameras. But the meetings confirmed what he knew: "Everyone does security the same way: They spend millions of dollars on their programs. The expectation is that if something bad happens you rewind the video." In other words, it wasn't having the kind of crime-stopping utility Shrestha envisioned.

At the same time, he was gaining confidence in his teachable video-scanning tool. It could identify when a human fell and got hurt, or when a weapon appeared. The software also could gauge how certain it was that a security incident occurred. Low confidence meant it would ping a member of's small team of humans to verify what was happening in the video. In cases of high confidence, it alerts a designated authority, such as a security chief on duty or local law enforcement.

Just because Shrestha trusted his technology, it didn't mean investors saw the point. "At that time, the venture community did not believe that physical security was an interesting space where you could build a venture-scale business," he says. There were dominant players already. Companies' budgets were allocated. But's solution was complementary with existing security: It could be integrated into almost any camera-feed system, and customized on the basis of the security needs of nearly any business to detect threats in real time. Still, Shrestha says raising the first $2 million for required approximately 50 meetings over the course of two months.

The company pitched its product where it saw immediate need. When a private school in San Jose, California, the Harker School, experienced a nighttime perimeter breach (caught on video that no one was watching) followed by an assault the next morning, Shrestha proposed his system could have prevented it by alerting the authorities immediately. Getting a paying customer seemed to set more deals in motion. While still in beta, the company slowly amassed a client roster. Investor confidence soared, too. When raised a Series A round of funding, it took 13 days of meetings; the Series B took just three.

After five years of signing up customers and building up its A.I. intelligence in stealth mode, formally launched to the public in January 2022. It also announced it had raised $52 million in a round led by Andreessen Horowitz. The startup works with seven of the top 10 U.S. technology companies by market capitalization, and its client list includes Adobe, VMware, and Impossible Foods. Most of the company's 100 employees are based around its headquarters in the San Francisco Bay area.

Shrestha is hoping his company flips the surveillance model of security to be proactive, rather than reactive. He's also addressing concerns about the use of machine learning in security, which evokes concern over baked-in or learned prejudices and profiling. The system identifies forms of objects and people, not their colors or traits. Unlike other video-monitoring systems, it does not use facial recognition. Nor does its system have the ability to recognize bias-inducing traits, such as gender, age, or skin color.

"It's not looking for classes that can include bias," Shrestha says. "There's a huge responsibility of people who build these systems to build systems from the ground up to maximize privacy and to eliminate bias."