I’m excited to share this interview with the newest addition to our team of AI data experts! Julia Liberty comes to LXT with a wealth of AI data expertise helping clients of all sizes with their data challenges. I sat down with Julia to learn more about her background and some of the interesting projects she has worked on in recent years.
Name
Julia Liberty
Role
Director of Business Development
Country of origin
Germany
Current location
United States
Years working in the AI data industry
5
Favorite AI application
There are many great AI applications, but I’d have to say ChatGPT has become a favorite. I use it almost every day.
Tell me a bit about your background. How did you end up in the field of AI data?
To be honest, my career in the AI industry was not planned. When I was in college in Germany studying language and communication with a focus on linguistics, I looked for a job that complemented my studies and started working for an AI start-up called Symanto AI. The company used psycholinguistics and people’s writing styles to build language models that analyze and predict consumer behavior so that companies can develop more targeted marketing strategies. Additionally, they built a large sentiment analysis model to underline positive and negative aspects of products, movies and consumer goods. At Symanto AI I had the opportunity to work on both the product and sales teams, which gave me great exposure to different aspects of the business.
When I moved to Seattle in 2017, I continued working for Symanto AI until I was recruited by AI Data Innovations as an Account Manager focusing on new business development. This was the start of my career in AI data, and over the past five years I’ve had the chance to work with a wide range of clients on all types of AI data projects.
I recently joined LXT and am excited to work with some amazing talents who’ve worked in the AI data space for years.
What made you decide to join LXT?
LXT’s language capabilities always impressed me; usually vendors cover a sliver of what LXT is now covering. Additionally, the team has such a vast amount of experience and talent in this space. I am excited to have the opportunity to learn from and work with them.
Furthermore, I was attracted to the fact that LXT has such a large global footprint and still manages to provide a welcoming work environment for employees. Many companies nowadays struggle with that and don’t pay enough attention to their employees’ needs. I’m excited to be a part of such a warm and friendly organization.
Over the years you’ve seen quite a range of AI use cases. What are some notable examples that come to mind?
The environment-focused use cases for AR/VR always come to mind. I worked on many data collection projects where our team created a variety of collection sites resembling real world scenarios, and some where we used actual real-world locations. The set-up for those projects and the teamwork needed to ensure a successful data collection outcome is quite extensive. Furthermore, those collections are often quite unpredictable when it comes to the equipment that is used (especially when using prototypes), the participants, and the project timelines.
I believe one of the main reasons that our team was so successful was that we were extremely organized and always planned for the worst-case scenario. For example, if a prototype wasn’t working we would engage directly with the client to understand what we can do on our end next time to troubleshoot without needing their input and use the overhead created by that issue to focus on tasks such as recruiting more difficult participants cells. We also learned what kind of team members are important to make an in-person collection successful; everyone on the project needs to be able to wear multiple hats – sometimes even at the same time.
Another example that comes to mind is an utterance generation project that I led. The client wanted a lot of variety, but at some point there is a natural point of exhaustion with the data. There is a balance between having enough variety in the data and ensuring that utterances sound natural, and I had long discussions with clients on how to get the variety of utterances they wanted while making sure that the utterances reflected real-world scenarios.
What is one of the most challenging AI data projects you’ve worked on?
Most data collection projects have unique challenges, particularly because they are all custom, and in-person collections are often more challenging. Outside of setting up the right environment and meeting the unique requirements of in-person collections, you must also make sure that you can meet client’s volume requirements. Many times the no-show rate is very high (depending on location, project type, collection length, etc.) and you need a strategy in place to ensure that your targets are not impacted by no-shows. Finding the balance between touching base with the participants often enough and when to overschedule is key. Additionally, having a team in place that is flexible is important, which allows you to quickly pivot and add additional collection days/hours to allow for more flexibility for participants.
In your role you help companies determine the type and amount of data they need to improve their AI solutions. How would you describe your approach?
My approach always starts with thoroughly understanding the customer’s end goal so that I design a solution to help reach that goal. Generally my clients have a high-level idea of the data they need, and my role is to help them think through the details. For example when it comes to data for Conversational AI, a client may want to collect utterances across genders and age ranges, but I encourage them to think more broadly and extend their collection into specific ethnicities, dialects, accents, and so on to make their product equally accessible for a wider range of users. It really depends on the target markets and customers they are trying to attract.
What advice do you have for companies working in AI when it comes to their data strategy?
I encourage companies to refine their requirements and make sure they reflect the real world. The right data partner can help define their data requirements to account for this. Then it’s critical to create thorough guidelines, especially when human participants are required. This reduces the potential for costly rework. Ultimately clients should do their due diligence and make sure that they pick a partner that is the right fit for their project to ensure the best outcome.
