Part I. Building AI-powered learning: a journey on how we used AI to enhance our learning experience
When ChatGPT launched in late 2022, it opened up an exciting new chapter for AI. At Preply, we saw an opportunity to do something special — taking this powerful technology and crafting it into tools that make a genuine difference for our tutors and students. Rather than rushing in, we took a thoughtful approach, focusing on creating features that enhance learning and connection. In this story, we take you behind the scenes of our early development stages, guided by Preply’s Senior Product Manager, Anna del Pino, who led the team pioneering the integration of AI into Preply’s platform.
“So, how did we start the journey? In November 2022, OpenAI and ChatGPT burst onto the scene, and suddenly, it felt like everything was speeding up. It was like a sprint — every company knew they had to jump in, but no one quite knew what the playbook was yet. It was exciting and overwhelming all at once. There was so much buzz around”, Anna recalls.
Cutting Through the Noise
Preply had a clear mission: cut through the noise and figure out practical, scalable ways to bring AI into our product. We weren’t just chasing the latest trend — we wanted to make AI a real game-changer for the learning experience. “Everyone was asking the same thing: ‘What do we do with this?’ For us, it was about figuring out how AI could really help both learners and tutors in a way that mattered,” Anna says.
“We began working on AI at Preply before we had a dedicated team — AI was an area of exploration within our core product team. Initially, we were at a crossroads with many questions ‘Which provider should we use? Will it be fast enough? How can we ensure high-quality AI-generated results?’ and so on. However, for me as a product designer, the main challenge and question was explaining the full power of AI for learning without making the interface overly complex.
Vlad Kyshkan, Product Designer
Our team was excited to jump in and get things moving quickly. ‘What’s great about Preply is that no one holds you back from testing new ideas that have strong hypothesis for customer benefit,’ Anna shares. With that mindset, we didn’t wait for a perfect plan. Instead, we took the initiative by focusing on real user challenges and seeing where generative AI could make a difference.
Identifying the Use Case: Personalized Learning
One of the first key insights was that tutors needed content specifically tailored to each student’s individual needs. Before, Preply’s platform couldn’t offer that level of personalization. But with generative AI, we saw a real opportunity to enhance the learning experience by creating custom exercises and teaching aids on the fly.
The solution? The team built a Teaching Assistant, weaving AI into the chat feature. This tool allowed tutors to easily generate personalized exercises. “Tutors could pick a topic, tweak the exercises, and send them directly to their students,” Anna says. “It was a straightforward feature, but exactly what they needed — personalized content, delivered effortlessly.”
The initial rollout was promising. Tutors quickly took to the teaching assistant feature, using it right away with hardly any promotion. Encouraged by this success, the team decided to expand its capabilities, responding to tutors who wanted extra support in teaching. But this didn’t quite land the same way — quality control quickly became a major issue — and we learned a valuable lesson. “We realized that to make sure the AI was truly helpful, we needed solid guardrails and top-notch prompts,” Anna explains.
“As a team, we faced significant challenges when we began working with AI. While we had strong insights into industry trends and emerging AI products, the biggest hurdle was figuring out how to apply AI to language learning. It required extensive exploration to first understand the specific needs of our students and tutors. Only then could we determine how AI could effectively address those needs and which problems it was best suited to solve.”
Andres Gregorio Puig, User Research Lead
Defining the Path for Self-Learning
In addition to solving for the tutor, we felt strongly that Generative AI could transform the student experience as well. Similar to our efforts with tutors, we first needed to better understand the relevant user needs. This led to more in-depth research into our learner’s journey on Preply, which guided us to opportunities to help our students between their lessons. We uncovered three key themes for what students were looking for while they waited for their next lesson:
- Enhance their vocabulary
- Improve conversational fluency
- Develop comprehension skills
We saw an opportunity to use ChatGPT to build a more dynamic conversational practice tool. “We realized we could really shake things up by offering an interactive, real-time conversation experience,” Anna explains. This insight drove the team to spend Q4 researching and crafting a strategy for what would become the TalkNow feature (one of the self-learning exercises we offer).
Since this was new territory for us, there were a lot of things to figure out — legal concerns, creative choices like tone of voice, naming, and even gamification. It made building the product pretty complex. Soon, we realized we needed a dedicated team to turn this vision into reality.
“The team’s major achievement that helped us see results faster was organizational rather than technical. Kudos to the team for pushing through the idea of building a working MLP. Initially, it felt like we were trying to do everything everywhere all at once. However, it was critically important to quickly demonstrate the feature’s feasibility. Once we cut the scope and created a simplified design for the MLP, we built it in about a week. Seeing that this was possible was a major breakthrough.”
Anna Gerdii, Front End Engineer
And this is really just the start 😉 With these early innovations, we started to lay the groundwork for AI-powered learning at Preply. But as we pushed to bring these innovations to even more people, new challenges popped up. So how did we tackle them and scale these AI-driven tools? In Part II, we’ll take you behind the scenes, sharing what we learned along the way and key takeaways from another Senior Product Manager, Emily Stott. Look out for the next chapter next week — you won’t want to miss it!
We are actively seeking talented candidates to help us build AI-powered learning that enables people’s progress. If you’re excited by this challenge apply to one of our open positions!
