My name is Mahla, and I am a second-year Master's student studying Linguistic Data Science at the University of Eastern Finland (UEF). This past summer, I had the wonderful opportunity to join Youpret as an intern focusing on AI interpretation. Over the past four months, I was able to combine my linguistics background with hands-on technology by deeply testing and evaluating how AI translates spoken language. Here is a look into my internship experience and what I learned along the way!
First, telling shortly, which were the main tasks and the goal of the internship?
My main goal was to test the Youpret app's AI interpretation feature. I started with the languages I know best—Persian (my native language), English, Italian, and Dari. For these, I spoke directly into the app and evaluated both the translation accuracy and the quality of the AI's voice. For languages I didn't know, I teamed up with friends and colleagues to help spot errors. If I couldn't find a native speaker, I got creative and played foreign news or podcasts, using my linguistics background and online tools to evaluate the AI's performance.
What did you find most interesting during the internship?
I honestly found working so closely with AI to be the most interesting part. Of course, I had used AI before, but I had never looked so deeply into the details of its behavior. During the testing process, I got to see firsthand the real differences between various AI models and how they each handle translation. It was also fascinating to explore different Text-to-Speech (TTS) models and realize just how much those specific settings change the final voice output and pronunciation. Getting to peek "under the hood" and understand these technical behaviors was a completely new and exciting experience for me!
Was there something that surprised you?
Yes! I was really surprised by how well SOMETIMES the AI handled cultural phrases and slang. Instead of giving a stiff, literal translation, it would sometimes adapt an Eastern cultural concept into something that sounded completely natural and acceptable to a Western listener, so it didn't sound weird at all.
What did you learn?
I learned so much about what happens "behind the scenes" of AI interpretation. I learned how to properly evaluate AI translation quality, how different text-to-speech models work, and how to design and analyze user feedback surveys.
Which were the "coolest" things regarding Youpret AI interpreting technology?
Testing it was just a lot of fun! Whenever it gave a perfect translation, I felt so excited, and sometimes when it made a funny mistake, it gave me a good laugh. But the coolest part is realizing how incredibly useful this technology is for short, everyday, low-risk situations. Human interpreters aren't always available in the blink of an eye. So, for things like calling a help desk after waiting in a long queue when the client only speaks their mother tongue, this AI is a fantastic, instant solution.
Were you able to apply the knowledge gained at university in practice?
Definitely! The internship perfectly bridged the gap between what I learned over the last two semesters at the University of Eastern Finland and the real world. Many of my university courses were quite theoretical, and seeing AI interpretation in action at Youpret really broadened my horizons. Now, when I speak to an AI, I actually understand how the neural machine translation systems are working behind the scenes!
How did you like Youpret as a work community?
It was honestly an amazing and unforgettable experience. My colleagues were so kind, helpful, and welcoming. At the same time, they gave me the freedom and trust to work, explore, and learn in my own unique "Mahla style"!
You were working closely with Faiza, who works as Junior Software Developer at Youpret. This is how Faiza described the joint project, testing the quality of the Youpret AI interpreter:
"We put AI interpreter through extensive multilingual testing across everyday conversations and real-world scenarios, including healthcare, social services, and general assistance.
The results were highly encouraging: translations generally preserved the intended meaning, audio was clear and successfully generated, and the end-to-end speech-to-text, translation, and text-to-speech process performed well across a wide range of languages.
The testing also highlighted important language-specific nuances, including differences in formality, terminology, writing systems, and occasional phrasing refinements.
All in all, the experience demonstrates the potential of AI interpretation to make communication more accessible, helping people overcome language barriers, understand essential information, and engage with services more easily—while maintaining appropriate human oversight where communication is particularly sensitive or high-stakes."
Mahla Baniasadi
Linguistic Data Science student at UEF
Summer Intern at Youpret