Tips for building an internal AI team

If this month’s MultiLingual theme isn’t enough of a sign, the translation industry is engulfed in the buzz of AI. We’re told this new technology is capable of disrupting everything about the way we work and live, and anyone not already considering how to integrate it is already behind. Companies around the world are scrambling to apply both the competitive edge and the buzzword to their business model — and you’re probably one of them, considering you’re here, reading this article.

The road to building an internal AI team isn’t as treacherous as you might think, says Dalibor Frívaldský, who built Memsource’s AI department from the ground up.

Everything can be summarized into a few simple dos — and equally important — don’ts.


Yes, the thought of having an in-house AI team is exciting, but it’s not necessarily the best business decision for all companies. Make sure you’re all in before taking the first steps.

Do put a lot of time and thought into whether this is really the right decision.

If cost and speed are your priority, it may be better to outsource. If you’re willing to invest both of those things, however, now is the best time to begin. The market for capable engineers and the race to implementing AI is only going to grow more competitive, and an in-house team makes more sense when evaluating long-term ROI if you’re willing to wait.

Don’t go into this without the proper support. This process will take a lot of work so it’s key to have the right people and conversations to ensure things go well. Assemble a team and make sure there is at least one person who explores the technology themselves. It’s vital to have an understanding of the technology before attempting to integrate it into your product, Frívaldský said. A quick Google search will reveal numerous courses on AI that can serve as your catalyst. Frívaldský recommends Andrew Ng’s class, “Introduction to Machine Learning,” to create a neural network capable of identifying images of cats. “And of course following the topic closely on whichever media suits you,” he suggested. The courses, combined with technology blogs, YouTube and Science Magazine provided him with enough of a knowledge base to find success in building a team.


Perhaps the most time consuming and definitely the most critical, where you choose to utilize AI will make all the difference.

Do focus primarily on your client’s pain points. Unless you’re using the technology to make someone’s life better, it will have a hard time moving beyond a buzzword. With this in mind, dive into your company data to identify logical use cases. In Memsource’s case, dozens of concepts were pitched before the perfect crossroads of user experience and existing data was identified. The team found that nontranslatables made up 14% of segments and 4% of content, and while nontranslatables is certainly not a novel concept, the application of AI to increase their detection is. Try to identify your own crossroads.

Don’t build AI for the sake of having AI. Approaching AI as a technical function alone runs the risk of falling victim to your own imagination, warned Frívaldský. Don’t get stuck on coming up with one big innovation — the sum of many smaller functions can still make a big difference to the user experience. Improving your existing product is the end goal. “That’s what AI is: a tool to make existing systems more efficient and accurate. You can’t pinpoint those areas of improvement if there is no existing system,“ he said.

You don’t even need to know exactly how you plan to apply AI before you begin the hiring process, but you should at least be exploring the problems you can realistically solve.


We’re in a hot market and building a highly specialized team is a hefty time investment. It’s going to take more than simply posting a job opening for an AI engineer.

Do recruit from academic institutions. Presenting at conferences and in related classes and developing relationships with institutions will go a long way. That’s how Memsource brought their first AI engineer, Aleš Tamchyna, on board.

Don’t get distracted by someone who is savvy with the highly technical jargon. Make sure that the person you’re hiring is capable of putting their words into practice. Frívaldský suggested issuing applicants a small AI or machine learning project as part of the application process, to be absolutely certain of their skills.

“Rather than focusing on getting the best performance from the project, look at the level of understanding the applicant shows as well as their code culture. Implementing some article goes a long way here, as the ability to work with state-of-the-art technology will be crucial down the line. Being able to understand AI articles and reproduce the results is an invaluable skill,” he said.


Is this the part where we dive into building industry-disrupting innovation? Well, not quite.

Do manage your expectations. AI is very different from standard development in one particular area — there’s no way of knowing if something is going to work or not until you actually try it. It is very exploratory; setbacks and failed attempts will happen and in fact will be quite common.

Not every exploration will be a success, but it will be one step closer to it. AI requires patience and a positive attitude.

Don’t try to reinvent the wheel. AI is new and exciting technology, but that doesn’t mean you have to dream up an entirely new process to implement it. Begin with an existing solution to a problem and consider how to further improve that solution. Following your intuition is a perfectly acceptable place to start, but don’t move forward until you’re sure the data proves you right.

Begin small and work your way up. Only after developing a better understanding will you be able to attempt new or previously unsolvable problems. 


Whether it’s your AI team or the AI itself, there’s always something new to learn.

Do maintain your data. In order to improve your AI’s capabilities, it needs to be fed high-quality, fresh and as close to real-time data as possible. Companies that prioritize this will quickly outpace the rest.

Don’t ever stop learning. Provide your AI and yourself new information as often as possible in order to grow and remain relevant in the exponential advancement of technology, as there will always be something new. “You never know when someone is going to publish an article that shatters the current state-of-the-art,” said Frívaldský. He sees this, however, as just another step in the evolution of technology. “That’s just the way this field works. It’s nothing to fear. It’s exactly what makes it exciting.”