CTO Sessions: Adam Sypniewski, Deepgram

Name: Adam Sypniewski

Company: Deepgram

Job title: CTO

Date started current role: June 2016

Location: San Francisco Bay Area

Adam Sypniewski has designed artificial intelligence systems for autonomous vehicles and built next-gen technology for The Defense Advanced Research Projects Agency (DARPA). He earned his Ph.D. in experimental astrophysics from the University of Michigan, where he used machine learning to understand dark energy.

What was your first job? I worked as a paperboy when I was eleven years old, making 10.2 cents for every paper I delivered. I saved all the money I earned so that I could buy my very first Compaq PC—you can only imagine how many papers it took before I had saved enough!

I started my first corporate job while I was finishing graduate school at the University of Michigan. I was looking for careers in software engineering and science, and there was a local company called Soar Technology. I applied and started working at the company part-time until I finished my Ph.D and then I transitioned to working there full-time. Initially, I was an AI Engineer, and eventually I became a Research Scientist where I worked primarily on unmanned and autonomous vehicle systems that used AI to help communicate with, or improve the effectiveness of, humans.

Did you always want to work in IT? I actually didn’t. Growing up, I enjoyed science, technology and electronics, and anticipated that I would pursue an undergraduate career in electrical engineering. I knew I liked programming, but I didn’t necessarily anticipate working in IT.

What was your education? Do you hold any certifications? What are they? I have a Bachelor of Science from the Alma College where I pursued a double-major in mathematics and physics and received honours in mathematics. I also have a Master’s degree and a Ph.D. in physics from the University of Michigan. 

Explain your career path. Did you take any detours? If so, discuss. Originally, I applied to the University of Michigan’s electrical engineering program but was turned down because I was home-schooled during high school; at the time, there wasn’t much support for students who were home-schooled, so I had to look around at other schools. I ended up applying and going to Alma College, a small liberal arts college in Michigan that had a 3 + 2 program—where I was able to go to school for three years at Alma College and then transfer for an additional two years of school at University of Michigan to finish with a Bachelor’s and Master’s in engineering. Alma College didn’t offer an engineering program, so I had to complete my undergraduate degree in physics until I could transfer to the University of Michigan for my graduate studies in engineering.

What ended up happening, however, is that I discovered that I enjoyed physics much more than I had ever enjoyed engineering. In fact, to this day, I don’t enjoy electrical engineering at all. I didn’t have a career path in mind, but I decided to follow my passion and continue my studies in physics. I received my Master’s in Physics from the University of Michigan and, upon completion, I applied to its PhD program. 

What type of CTO are you?  I’m a big fan of taking new or “risky” technologies and trying to think about: does this change the way we think about things, and can it do what we need it to? If so, does it push one’s assumptions about the universe or is it a hard constraint that we can’t push up against? As a CTO, I am very fond of taking a chance on the so-called riskier endeavours and technologies that can help us as a company optimise the way we work in a constraint based world. 

Which emerging technology are you most excited about the prospect of? I think that the work being done in development and silicone is very interesting. We have CPUs, GPUs, FPGAs, as well as more refined machines for doing targeted specific calculations quickly. The more general the tool is for a particular job or task, the less efficient it will be on that same, particular task. This is because these technologies are designed to solve a lot of problems and so they have more general instructions. GPUs for example, are more specific, and are designed to tackle a specified issue incredibly well, because that is all they are designed for. Specialised computing opens up many doors in order for companies to do things cheaper and more effectively.

I’m also excited by the prospect of nuclear fusion as a power source. For example, how do we push the boundaries of space and what people can do? Nuclear fusion provides an inexhaustible supply of energy. It’s an efficient technology and I think it will relieve a lot of pressure on the energy market.

Are there any technologies which you think are overhyped? Why? I think that blockchain is overhyped. At its core, these are new applications of existing technologies, but people are trying to shoehorn blockchain wherever they can possibly find it. Rather than thinking about the actual applications of blockchain, I think that people are caught up on the idea of blockchain as a buzzy trending technology and often I see it used as a marketing buzz word or a way to entice investors.

Artificial intelligence (AI) and machine learning (ML) can often be overhyped as well. At Deepgram, we have a whole research branch and are dedicated to making AI and ML part of our company DNA. Oftentimes, though, AI and ML can be used as a marketing buzzword for new companies looking to remain competitive and it can be hard to differentiate which companies are putting in the time and research and energy to develop these types of emerging technologies and which companies aren’t putting in the necessary work.

What is one unique initiative that you’ve employed over the last 12 months that you’re really proud of? Over the last twelve months, we’ve made an all-in commitment to using safer programming technologies, specifically the Rust programming language. We’re using it for speed, low-level computation and to drive our GPUs. Very few companies are doing this at all, let alone in Rust, and it has made a huge impact on the stability and speed of our systems. As programming languages, Rust and Elm have a strong emphasis on correctness and are regarded positively amongst programmers.

At Deepgram, our front-end technologies (for example the UIs we build in a web browser) are frequently built in Rust and leverage Elm as well. This shows Deepgram’s interest in and ability to adopt novel technologies, even when they’re not widely adopted across the industry. Over the last twelve months, I have seen extraordinary excitement by the team at Deepgram about our use of Rust.

Are you leading a digital transformation? If so, does it emphasise customer experience and revenue growth or operational efficiency? If both, how do you balance the two? No, not necessarily. We’re actually moving backwards. What I mean by that is that we are physically going out to data centres with real GPUs that we at Deepgram are physically maintaining, rather than investing significantly in cloud services. If something goes wrong with the equipment, one of our engineers will take care of it.

What is the biggest issue that you’re helping customers with at the moment? At Deepgram, we believe that voice is the future of the enterprise, and we help companies capitalise on their audio data. We use a patented deep learning model that allows companies to get faster, more accurate transcriptions, resulting in a more affordable and more scalable path to releasing new speech-driven products. 

How do you align your technology use to meet business goals? Deepgram is a foundational technology company. We provide a fundamental API that customers can use for speech recognition. All of the business we do is centred around the technology that we have built. Our speech recognition technology provides the right solution to their problem—companies turn to Deepgram looking to uncover actionable insights hidden within their speech data, Deepgram offers an easy, affordable and scalable solution.

Do you have any trouble matching product/service strategy with tech strategy? Typically, that sort of friction comes up when there are unrealistic or naive expectations about how the technology will work. You have to be careful when you’re in a “deep tech” company, or when you have interesting and novel technology, because people will make all sorts of assumptions about how it works, what can or cannot be done, how easy your solution is to use, etc. To avoid these issues, your internal communications have to be really good. The way that you describe your own product solution has to be very honest so that your product and marketing team can work together to build and market exactly what you’re working to achieve.

You can also see friction between the product and tech strategy when you try and think “too big”. You want to be laser-focused on the value-add that you bring to the market. It’s easy to get distracted from that initial focus, especially when building a product, because you want to keep building and thinking about the next big thing that your customers are demanding. But, you also have to learn to say no and focus on the core proficiency of your company and first ensure that you’re doing that exceptionally well. 

What makes an effective tech strategy? I think that it’s important to be able to forecast the technology or solution that is going to make a difference and then challenge the assumptions of what can and cannot be done. Ask yourself why your particular tech strategy will make a difference and why it hasn’t been done before, and go from there.

What predictions do you have for the role of the CTO in the future?  I think the CTO role will become more focused on using the right tools for the job, and less focused on picking the right off-the-shelf solution. CTOs will be more focused on making strategic technology-focused decisions, looking at how the market is going to change, and what that means for the technology that you should be adopting now. Customers will also undergo similar technological shifts, so it’s critical to anticipate what your customers will need or will likely ask for down the line and build that technology now in order to support them better.

What has been your greatest career achievement? I would say that building Deepgram’s speech engine has been my greatest career achievement to date. We built several versions, and it didn’t happen overnight. Ultimately, Scott Stephenson (the CEO and co-founder) and I discovered that the tools available to us just weren’t working and we needed a better solution.  

Looking back with 20:20 hindsight, what would you have done differently? Very little honestly. Things worked out so well that, by and large, 80-90% of the decisions we made were exactly right. With the knowledge I have now, I likely would have made the same decisions. In terms of what I would have done differently, I would have started a data operations initiative sooner. We implemented one several years ago, but if we had made the decision 6-12 months earlier, we would have benefitted from it 6-12 months sooner, and that would have been really big for the company.

Also, we were slow to “throw it out and start again”. Looking back, it would have been beneficial to recognise when we needed to start over from scratch and not try to continue to make something work.