When ChatGPT emerged at the end of 2022, my startup was in the midst of seeking a transformation. After exploring over a dozen directions, I finally chose to focus resources on an AI speaking practice product, TalkBerry. A major pain point in speaking practice is the lack of practice context and a practice partner. LLMs can not only simulate various practice scenarios but also act as virtual practice partners, so at the time, I believed AI speaking was an excellent application scenario for LLMs.
Early Feature Exploration through a Chrome Plugin
To test the waters and understand our potential users, we moved quickly. In early 2023, we developed and launched a Chrome plugin called "Talk to ChatGPT" in just few weeks. The concept was simple: it allowed users to interact with the ChatGPT web interface using their voice.
The response was immediate and overwhelming. Within a short period, our plugin surpassed 20,000 installations. We were featured on prominent tech sites like LifeHacker, which drove a significant amount of attention to our project.
Home page of TalkBerry Chrome Extension.
Covered by LifeHacker
This initial experiment was invaluable. It connected us with a passionate community of language learners who were already using ChatGPT, helped us understand their core needs, and funneled tens of thousands of users to our official website each month through organic search traffic.
Design the Core Experience: The TalkBerry Website
Armed with these insights, we dedicated the next two months to developing a standalone web application. Our goal was to create an AI-powered speaking app that was more than just a chatbot. We wanted to provide a structured and immersive learning experience. Key features included:
Multiple Learning Modes: TalkBerry offers a range of dialogue formats including virtual characters and everyday scenarios, allowing users to jump straight into realistic speaking practice. It’s more convenient and effective than using ChatGPT directly.
Structured Courses: We’ve organized different scenarios into thematic courses, enabling users to focus on specific topics. The AI also provides targeted feedback to help users improve their speaking skills in a more structured way.
Guidance and Feedback: Throughout the conversations with AI, we offer various learning aids such as translations, prompts, and AI-generated feedback, allowing users to learn as they practice.
Listening to Our Users: Key Lessons from the Web App
After launching the web version of TalkBerry, we intentionally avoided paid marketing. Instead, we focused on organic growth to validate the product's intrinsic value. This strategy brought us nearly thousands paying users each month and, more importantly, yielded critical insights that would shape our next steps.
Mobile is more aligned with real usage scenarios: Although TalkBerry was built as a web product, over 50% of our traffic came from mobile devices. Through user interviews, we found that 85% of users preferred to practice speaking on their phones anytime, anywhere, rather than being tied to a computer. This insight led us to shift our focus toward mobile moving forward.
Large language models can’t fully replace structured courses, but they greatly enhance the learning experience: Initially, we believed that LLMs could replace traditional curricula and content. However, user behavior showed that people still need a structured syllabus and courses to provide a knowledge framework and clear learning path. AI’s real value lies in unlocking new ways of learning and making content creation much more efficient.
Launched TalkBerry for iOS
Guided by these lessons, we shifted our focus to mobile and developed TalkBerry for iOS. This version was our most polished yet, featuring more structured courses, a wider array of practice scenarios, and more precise speaking feedback. The app was well-received, and we grew to several hundred paying users.
However, despite this progress, the revenue generated was not enough to sustain our four-person team. We faced a difficult choice, and ultimately, we had to make the tough decision to halt further development of TalkBerry.
While the journey of TalkBerry has come to an end, the experience was a profound lesson in product development, user feedback, and the dynamic landscape of AI. We are proud of what we built and deeply grateful to the community that supported us along the way.