Imagine a digital language tutor who greets you with a friendly “Hey, bro!” and adapts each lesson to your skill level as you learn. Not long ago, this sounded futuristic, but breakthroughs in large language models (LLMs) and lifelike voice cloning have made it a reality.
In this article, we’ll take you behind the scenes of our collaboration with Kuba English—an AI-driven language app that brings a personalized tutor straight to your pocket. Built on GPT models, the app doesn’t just sound like Kuba; it captures his laid-back teaching style, creating an interactive experience that feels truly alive.
Let’s explore the key challenges, solutions, and insights that helped shape it.
The challenge: Building an app that feels alive
Creating a personalized English-learning app goes beyond assembling lessons and quizzes—it needs to be engaging, adaptive to user needs, and true to the unique personality of its teacher, Kuba from Kuba English.
We spent a lot of time going over prompts to make sure the conversations sounded natural and stayed true to Kuba’s tone, without feeling too scripted or generic. Using LangSmith’s playground, we tested different conversational scenarios—structured lessons and casual chats with role-playing exercises—refined the prompts, and adjusted model parameters to keep the AI tutor’s responses authentic and engaging.
Another challenge was keeping the outputs consistent and finding the right balance between being too strict and too lenient in the app’s feedback. During development, we noticed that the AI sometimes corrected users too harshly or overlooked important areas for improvement. To address this, we used structured outputs from GPT, defining a clear response format.
Technical implementation: Building the brains and voice
AI avatar with signature voice
One of our earliest goals was to create an AI avatar of Kuba that’s more than just his name and face. Using ElevenLabs’ advanced voice cloning technology, we recorded and processed Kuba’s speech to capture his natural tone and emotional range. The result is a lifelike reproduction of his voice, with natural intonation, adding warmth and personality to the AI tutor.
Conversation logic
At the heart of the app, GPT models from OpenAI act as the “brains” of each conversation. They analyze the context and the user’s messages to make sure every response feels natural, supportive, and consistent with the tutor’s friendly teaching style.
Lesson structure and personalization
As a student works through the lessons, the AI tutor makes the language and tasks more challenging to keep them engaged. After each lesson, the app reviews the conversation and suggests extra content—like podcasts or additional lessons—based on the student’s mistakes or areas of potential improvement.
Freestyle chat and role-playing
In addition to structured lessons, users can have freestyle chats to practice real-life conversations. They can talk to different characters—a strict teacher, a wise philosopher, or even a six-year-old child—each infused with Kuba’s signature humor and style. Students can choose from prepared topics like travel, nightlife, sports, and movies, or select their own.
Most importantly, the app encourages speaking practice: users can talk to the AI through voice input and hear Kuba’s voice in response. This helps tackle one of the biggest challenges in language learning—speaking fluently.
Real-time feedback
When students need help, the app gives them relevant suggestions to keep the conversation going. If something is unclear, it also provides translations with examples to show different meanings.
We also implemented a targeted feedback feature. At any point, users can ask the AI tutor to review their latest message. GPT then highlights mistakes or praises a flawless performance.
After each lesson, the app provides a detailed review, summarizing the student’s key achievements and recurring mistakes, and offers tips for improvement—like “Try to include a conditional phrase”—all in a friendly and supportive way.
Structured outputs
To make feedback balanced and useful, we use structured outputs from GPT. Typically, the AI-generated text varies slightly with each response, which is fine for casual conversations. However, when we need consistent outputs for further processing, unpredictable variations can “break” the code and become a challenge.
That’s where structured outputs come in. By defining a fixed format—such as correct/praise/suggest for feedback—we ensure consistency across the app and UI elements. This approach not only guarantees reliability but also transforms feedback into actionable insights for future lessons or extra practice.
Combining corrections with positive encouragement keeps students motivated and continuously improving.
Beyond language learning: The possibilities of AI
The Kubova English app is more than a testament to what AI can do—it shows how technology can create a learning experience that feels personal and truly human. By combining GPT’s real-time language processing, ElevenLabs’ voice cloning, and carefully engineered conversations, we’ve made sure the app embodies Kuba’s unique teaching style.
This mix of interactive dialogue, tailored feedback, and lifelike voice not only guides students but also encourages them to speak confidently, make mistakes, and keep improving.
Looking ahead, the possibilities for AI in education—and many other areas—are boundless. From personalized healthcare and more empathetic customer service to data-driven solutions in finance, AI has the potential to change the way we solve problems and connect with others.
At Applifting, our goal is simple: to keep pushing the limits of technology to create experiences that are meaningful, inclusive, and focused on people above all.
