Join Some of the Brightest Minds in AI
Turn cutting-edge ML techniques into enterprise-ready AI solutions for the world's largest organizations
We asked Kristian Tashkov, a Senior ML Engineering Manager who joined our Sofia office back in 2016, about what a day in his life looks like now, his favorite part about working at Hyperscience and more.
In the distant 2016, I was finishing my master’s degree in Artificial Intelligence and preparing for a summer internship at Google. I was really interested in everything Machine Learning from my classes, and having just quit my previous job at yet another big company, I was looking for something different. Through talking with friends, classmates, and even employees at Google about interesting start-ups using ML in Sofia, I discovered Hyperscience!
Just reading about what they were doing on their website (which wasn’t much back then) got me excited, but it was the interview process that won me over. The interview was challenging but also fun – I loved discussing complex problems with all the smart people at the company.
Four years later, that hasn’t changed, and working at Hyperscience means solving challenging problems with some of the smartest and most fun people I have ever worked with.
Engineering Managers at Hyperscience have a variety of responsibilities, which is what makes every day so exciting and diverse. Most of our daily standups or planning meetings happen later in the day when people in the U.S. wake up, so the first half of my day is syncing with my team and trying to see if I can help with what they are working on.
In Machine Learning there often isn’t a single correct solution for a specific problem, so you must experiment until you find the one that works. This means we are often brainstorming ideas with the team and discussing next steps on specific experiments.
Some days we have scheduled board game nights or happy hours on Zoom. It’s not the same as seeing your coworkers in the office and chatting in person, but given the circumstances, it’s a really good alternative.
Documents often have a predefined template (e.g. a “New Account” form for a bank that always looks the same, but the values people fill in are different), but you don’t know which one it is when the input is a filled in, scanned image. I managed to try different approaches from a variety of scientific papers and ultimately improved upon them for the Hyperscience Platform, delivering significantly increased accuracy and recall. This code is being used to automate documents in production daily!
Since becoming an Engineering Manager, this type of challenge and reward has only increased. I’m able to collaborate with my team on multiple projects every release, such as the one that I’m currently tackling which is automating complex tables on non-templateable documents.
Most of the work in Hyperscience is done in Python and the ML organization uses libraries like Pytorch, Numpy and OpenCV daily.
While ML isn’t a new concept, it has exploded in popularity in recent years and seeing how much we can push the capabilities of AI has been extremely exciting. We stay in touch with recent developments in the area, and our models are pushing the state-of-the-art when it comes to document processing.
The work is always challenging and exciting. Usually after two years, I start to feel myself wanting something novel and compelling.
After four years at Hyperscience, I can honestly say that every new release brings a new challenge. This variety keeps me on my toes and is amplified by the great team we have in place to help solve these challenges.
Try to understand how different approaches inner workings function on top of knowing when and how to use them. At Hyperscience the out of the box is usually just the start, but without adding novel ideas and improvements, it is never enough as a final solution.