Meet Ted Kyi, VP of Data Science

As Deep Sentinel expands and enriches our talent pool, we’re introducing some of the bright minds that bring you the best protection in the security industry. Another new face to the Deep Sentinel team, Ted Kyi is no stranger to AI and machine learning, and we’re thrilled to have his expertise as our VP of Data Science.

Ted studied computer science at Princeton and Berkeley (decent schools, so we’ve heard) before jumping headlong into the startup world. He has worked for several successful startups over the last 25 years, including a software company he founded. Ted is also an active member of the machine learning (ML) community and participates in mentorship programs. He lives in San Diego with his wife and two daughters.

Get to know more about Ted below!

Deep Sentinel: What brings you to Deep Sentinel?

Ted Kyi: I met Selly online, probably at least three years ago, because I’m very active in the ML community… And then one thing led to another.

I’m excited to be coming to a company that embraces AI. My last position was in the healthcare sector, where most organizations are far behind other sectors in adopting machine learning and AI technology. At Deep Sentinel, AI is a fundamental building block that powers up our guards to help many more people than they could without the technology.

I see a lot of opportunities to level up the product and add functionality. We can also use machine learning to improve other areas of the business, such as guard performance and marketing.

Deep Sentinel: Where do you see Deep Sentinel going? What excites you about our future as a company?

Ted: I read a lot about AI, and sometimes people debate whether AI can replace people for various tasks. While we may have answers to those debates soon for some jobs, and years from now for others, one thing seems clear: using AI to augment how people perform tasks is within our reach right now.

The pattern of using AI with humans in the loop is a great use case. A perfect example is how Deep Sentinel guards are informed by AI but ultimately use their judgment to evaluate every situation. I see many opportunities to expand our analytics and further enhance our ability to protect people and their property.

There are two sides to machine learning and AI when I think about Deep Sentinel. There’s the core building block part that’s part of the product. It would be very inefficient to hire enough guards to watch every camera for every customer. They have to watch only the interesting ones, and the AI decides when something interesting is happening.

But there are a lot of other use cases. For example, we can run analytics on people who come to our website. Another example is that when we do interventions, you could get an email to say, “We didn’t need to call you, but there was some guy, and we pushed him off before anything bad happened.”

Right now, there’s a lot of focus on the core building block side. But beyond that, are we adding more features? Are we also helping the company be more efficient, smarter, and more automated?

Deep Sentinel: What topics or fields do you specialize in? Can you give examples from your past?

Ted: I consider myself a machine learning generalist, so I stay in touch with the largest advances across a broad swath of machine learning. Recently, it’s impossible to follow AI without noting the tremendous advances in deep neural networks, specifically the transformer architecture. I like to read papers about the mechanistic interpretability of large language models.

To be honest, most of the projects I’ve done don’t sound terribly exciting. It was just like, “Oh, figure out if this person’s going to schedule an appointment. Figure out this person’s going to cancel.”

We had this one system that was identifying if you know certain things about a patient’s medical history. But you don’t have access to the list of all their diagnoses. Can you infer what diagnoses they have?

We ended up going with an expert system, as opposed to what we think of as an AI-type system. Instead of getting a bunch of data and training the model, we talked to a cardiologist. They said, “If this person has this diagnosis, we’re going to prescribe medication A and medication B. So if you see that combination, you know they have this particular condition.”

It was a lot of those things and nothing too terribly cool. There are just so many mundane uses of machine learning.

Deep Sentinel: What will be the most exciting thing about AI and machine learning in the next, say, three years?

Ted: People were generally aware of AI, but it just wasn’t that wow until a little over a year ago when ChatGPT came out. Now it’s no longer something engineers are going to build into a product. This is something that I can see, feel, touch, and use myself.

The layer that I would add is that language has been designed, evolved, and improved upon by humans for communication. We use language to communicate everything. If there’s a concept and they don’t have a way to describe it, they’ll just create new words. The net result is that we have language to describe every concept we possibly know about.

And this means language can be the universal translator for anything. You can use language to describe a picture, music, or what’s happening in a video. You can language to describe somebody’s voice or tone or emotional state or the expression on their face.

I say all this because now we have very strong language models—and nobody’s saying they’re perfect. Last year was really about this explosion with language models starting with ChatGPT. This year, the focus is on multimodal models.

Multimodal models will be different. Not just because it’s really nice that you can do both text and pictures. No, I think it’s actually bigger than that because language can glue anything together.

If you have a model that understands video and another model that understands music, you can say, “How would you score this movie? What song would you pick for this scene?” Because even if there is no video-to-music translator, you can always use language as the in-between.

That’s the beauty of language. You can glue anything together. If people aren’t yet blown away by AI, this will take it to the next level.

Deep Sentinel: What do you like to do outside of work?

Ted: Well, this isn’t far outside of work, but after hours I am active in the San Diego machine learning community. I lead ML book clubs and paper reading groups.

I also participate in data science competitions. Last year, my team won an international competition about finding where the ink was located on Roman papyrus scrolls that were carbonized when Mount Vesuvius erupted in the year 79 AD. That competition kicked off a yearlong effort which ended with the successful decoding of several passages.

You can read about how the deciphered scroll discussed Epicurean philosophy, was likely authored by Philodemus, and the technology behind this work at https://scrollprize.org and this cool Scientific American article.

I will say that if I weren’t doing what I’m doing, one thing that I might like to do would be to teach. That’s why I do mentoring.

Deep Sentinel: Do you have any additional thoughts you’d like to share? How about your favorite sport?

Ted: I enjoy baseball more than any other sport. I admit that my love of baseball partly comes from the long, rich history of baseball analytics that is intertwined with the sport.

But another reason why baseball can be so much fun is because it is one of the only sports where you can still see people who look like you and me being highly successful. I love a good underdog story, and it’s cool to see people who don’t look like Superman become stars.

[Editor’s note: For inquisitive readers, Ted grew up in the New York area and is a lifelong Yankees fan.]

Need a Solution that Prevents Crime? Deep Sentinel is the only security technology that delivers the experience of a personal guard on every customer’s home and business. Call 833.983.6006 for your free security consultation.