Employee Spotlight: Alexander Kimiavi

Hi Alexander, as one of Dedrone's precious software engineers, can you tell us a bit more about your day-to-day activities?

My day-to-day life at Dedrone can vary and it involves many things, but I start the day with the computer vision daily scrum. This gives our team time to sync up on our projects, enabling us to tackle road blockers and ask questions, ensuring that we are operating as efficiently as possible.  

Next, I spend part of my day planning future projects and research plans for the computer vision team. The issues we tackle evolve daily, so it is important to effectively plan what research we will be doing as a team to address these problems. I then spend the rest of my day researching our models or developing code to enable the research of the models. This involves developing and researching our neural network architectures, making data driven decisions on what to sample from our dataset and what data to curate, adding features to our video software, and increasing the efficiency of our video inference.  

What do you like best about working at Dedrone?

The best part about working at Dedrone is the people I interact with on a daily basis. Everybody at Dedrone has their own area of expertise, and it is exciting to learn from them and their respective fields. Furthermore, the problems we get to solve at Dedrone are challenging which makes the work fast paced and exciting.  

Tell us about the best assignment you’ve had working at Dedrone and why it was so awesome.  

The best assignment I have had here at Dedrone was spearheading the research and development of Pythagoras 1. This required a complete overhaul of how we do computer vision at Dedrone. This overhaul involved upgrading how we source our data, engineering new and efficient pipelines to handle millions of images for training, researching a brand new in-house neural network architecture, and lastly, leveraging cutting edge technology to accelerate our inference on the edge.  

The research and development were fast paced and involved us leveraging the bleeding edge to develop this model. Seeing this model come to life and the improvements it brought to the product was very fulfilling and exciting.  I am incredibly proud of what we accomplished in a relatively short period of time.

What do you think is the most important element of airspace security that people might not understand?  

I think people fail to realize the true scope of the problem. The air is filled with a lot more targets than you would expect. To detect these, we use a wide variety of sensors that all have their own error associated with them. Different sensors may also detect the same target—these must be fused together algorithmically to provide a clear picture of the airspace. In terms of computer vision, there are a large variety of aerial targets and scenes that your detector needs to be robust too. To address these challenges, you need a strong fusion engine and a great machine learning team which, thankfully, Dedrone has. The problem is not a simple one, and it is constantly evolving.

Outside of your work with Dedrone, what do you enjoy doing in your free time?

I really love the work that I do, so in my free time I enjoy keeping up with the latest research in deep learning. I also enjoy working on side projects, going for long walks and exercising, and socializing with friends! When I can, I love helping my parents take care of their sweet dog Ivy.

‍