Deep Tech Companies in Armenia: What Problems Do They Solve?

Illustration by Armine Shahbazyan.

What is a Deep Tech company?

Deep Tech companies create breakthrough technologies to solve the problems of the modern world based on new knowledge and engineering innovations created by scientific-research groups. In other words, specialists in different fields—scientists, engineers, entrepreneurs—unite to solve a complex and fundamental global issue, creating the technology based on research and innovation that will solve that issue.

The solution to a global issue proposed by such companies can change the status quo economic situation—create a new market or destroy existing ones. This is the case with disruptive innovations. Deep Tech technologies—artificial intelligence; augmented and virtual reality; blockchain and cryptocurrencies; Internet of Things; autonomous machines; quantum technologies; biotechnology; 3D printing—will be the driving force of the global economy for the next decade: dubbed the fourth wave of world-changing innovation.

We have singled out three Deep Tech companies operating in Armenia—Krisp, Unum and Denovo Sciences—to show what issues they solve, why it is important for Armenia to develop a Deep Tech ecosystem, and what steps need to be taken to become a country that solves ambitious issues and adds value to the world.

Krisp: Filtering Out the Noise

A neighbor making repairs during a work call, a car alarm, children playing in the yard, or even a family member having their own conversation no longer need to be issues for remote workers during their conference calls if they start using Krisp. Noise filtering was Krisp’s first product; the prototype had taken six months to put together. The app was released two years later. This is a fairly fast pace for a Deep Tech company; it can often take years of work to solve a problem.

The head of the company’s scientific-research group, Tigran Sargsyan, explains: “We facilitate holding more effective meetings and being more professional during that time. According to Maslow’s hierarchy of needs, the basic human needs are of a physiological nature. When conducting online meetings, being quiet and focused becomes a basic requirement. Krisp solves this very problem.”

Krisp is currently working in more than 10 different research areas. In particular, for the last year and a half, they have been working on separating voices, creating a technology that can differentiate the voices of two people speaking simultaneously, and transmitting only one of them. The prototype has already been created, and preparations are being made for production.

“Research projects are unpredictable. We can work in a certain direction for three months, realize the problem remains unsolved, sit down and discuss things again, decide on a different approach, and go in that direction for a few months. But we may then realize that this also does not work, try a third option and succeed,” says Sargsyan. “That is the nature of research. It is not a production line where we can set a deadline to reach from A to B. We might work for a month and then realize that it is going to be a year-long effort, or we might find an interesting solution very quickly.”

Turning conversations into text is an example. It would be useful for a person who missed a meeting to receive the text of what was discussed, so they can quickly go over it without watching the entire recording.

Voice separation technology is another challenge, that is, recognizing what is being said by two different people talking simultaneously, which is different from separating one voice from background noise. Krisp’s beta version for such a product is ready. For comparison, Google has been working on this issue for three years, but has not yet released a usable product. Sargsyan shared that they are now working on the technology to separate voices when there are more than two people speaking at once.

Guessing emotions is another challenge, understanding whether a person is happy, sad, angry or neutral. This can be understood by combining the voice, facial expression, the words used and context. Krisp aims to become an AI communication assistive technology, with the ability to count how many times a person hesitated (said umm) during their presentation, stammered, or how long they speak during meetings, how many times they interrupted others or were interrupted by others, etc.

Another research direction still in the initial stages is Beautification, but a demo result is already available. It entails mapping 3D points of one’s face—that is, tens of thousands of points—and identifying the lips, nose, eyes, etc. Once this mapping is complete, software could theoretically make changes. For example, if a person has not slept well and has red eyes, this can be corrected in the video in that their counterparts in the meeting see. Teeth can be whitened, skin can be smoothened, virtual lipstick applied, all in a realistic way.

Krisp’s research team consists mostly of scientists with PhDs in physics or mathematics, but who have also learned the components of programming and machine learning.

Krisp creates technologies that Google, WebEx, Microsoft Teams and Zoom also have. The difference is that the Krisp app runs on your computer as a virtual microphone, speaker and camera, removing background noise in both directions while a person speaks.

The resources of Krisp’s competitors are unlimited. “If we were to compare, they may have 100-300 more teams,” Sargsyan explains.” If we have to think about what criteria to set in order to get results at once and be able to test it several times, they can do a thousand tests in parallel, and choose the best result.” 

While this may seem daunting, Krisp’s teams focus on a certain problem and invest all their potential in solving that problem. And that is what a Deep Tech company strives to do – find problems and then create solutions to make people’s lives easier and processes more efficient.

Sargsyan says that many problems have been solved and what is needed now are teams that can solve the very difficult problems. To do that, new approaches are necessary, however, in Armenia, there are limited specialists. “Companies are looking for them in the daylight with a flashlight to find them, to persuade and hire them. From that point of view, children who are currently excelling in mathematics can be sure that they will have a bright future,” says Sargsyan.

Krisp employees are also shareholders of the company. In addition to high salaries, the money generated from the shares could also potentially stay in Armenia, if these employees decide to invest their own capital in new startups in Armenia.

UNUM: Revolutionizing Data Processing

Imagine that all the memories of your life are available with as much accuracy as this line you just read. You remember the content of the book you read at the age of 5 from beginning to end and can reproduce it without any mistakes. Now try to transfer this idea to the domain of computers. Computers have both long-term and short-term memory. Short-term memory is very fast, but more expensive and therefore smaller. Long-term memory is slower, but cheaper; it can be scaled up to store much larger amounts of data that you may not need to access very frequently. All of today’s complex data processing is done with short-term memory.

UNUM is trying to move those operations to the computer’s long-term memory. If they can make it happen, it will be possible to work with exponentially larger datasets at once.

The company was founded in 2015 in Yerevan as a private Deep Tech research laboratory focusing on the design of next-generation AI models. All the multidisciplinary work done in previous years by founder Ashot Vardanyan came together to create UNUM.

Vardanyan visited dozens of countries before creating UNUM, and participated in numerous professional conferences to get acquainted with world-class scientists and specialists. Now, he aims to “connect” their knowledge and resources with Armenia. In 2020, he moved to Armenia, gathered his team and continues to work to develop UNUM.

UNUM is one of the platforms offering scientific-innovative solutions in Armenia. It competes with global companies such as Oracle Corporation, SAP, MongoDB, Inc., Snow Software and Elastic NV. UNUM’s work domain is artificial intelligence, but since the infrastructure is not in place to achieve the ultimate goal set by the company, they write and create the necessary infrastructure from scratch. To solve the problem efficiently and much faster is the model that the company strives for.

Denovo Sciences: Using Algorithms to Conquer Disease

Denovo Sciences was founded in December 2020, though the team was already working on the algorithm within the framework of the 2019 FAST ASCENT program. The company employs seven people who are an interdisciplinary group of specialists: machine learning, computational biology and chemistry. They are trying to create small molecules of therapeutic value using machine learning algorithms created by the team.

Some of their competitors (ex. Insilico Medicine, Benevolent AI, Iktos) had a head start and are better plugged into financial and human resources. But Denovo Sciences is trying to pioneer new approaches and present them in an interesting light, says the company’s founder Hovakim Zakaryan.

“We use RL[1] to create small molecules that can interact with RNA targets. In addition, our algorithm allows us to create molecules that can interact with two different targets at once, thus increasing the therapeutic value of the molecules. These are new types of molecules that, if successful, could be the key to treating a number of diseases, or make existing treatments more effective,” explains Zakaryan.

According to Hovakim Zakaryan, opportunities will be created when scientific and education funding is dramatically increased.

“First, the state needs to facilitate state-of-the-art conditions for scientific work, so that it becomes possible to have competitive laboratories in Armenia and generate new knowledge. Next, the state should build a legal environment that will encourage private investment in such companies. And finally, the state must also participate in the creation and stabilization of venture funds, both by adopting laws and providing investment capital. And for all this, of course, we need human capital and resources,” emphasizes Zakaryan.

The most important challenge the ecosystem faces is the lack of scientific thought. Deep Tech companies are a deep ocean of opportunities. Often, new research branches are added during initial research, which in turn can become solutions to new problems down the road. The risks are great, but so is the profit—if we are successful.

1. Reinforcement learning is an area of machine learning.

Leave A Comment

Your email address will not be published. Required fields are marked *