Duonosaur

mainpage
loading page
dinolingo
whack-a-mole
chatbot teacher

Inspiration

Duonosaur was born from a very honest look at today’s language-learning apps—especially the ones we all use and love, like Duolingo. As frequent users, we realized that while these platforms are incredibly popular, they often struggle to keep learners engaged over the long term. Exercises can feel repetitive, pronunciation practice tends to be very surface-level, and grammar instruction rarely connects to how people actually speak in real life.

We asked ourselves a simple question: Why does language learning so often feel like a chore? From that reflection came the core idea behind Duonosaur—an experience built around active participation, instant feedback, and meaningful immersion. Instead of memorizing words or tapping through the same patterns, learners should be playing, experimenting, and getting real-time responses that help them improve. That vision shaped our mission: to blend education and entertainment so seamlessly that users don’t feel like they’re “studying,” but genuinely enjoying the process of learning a new language.

What it does

Duonosaur is a fully interactive language-learning ecosystem designed around two game modes and an AI chatbot, each targeting a major pillar of language mastery.

Dinolingo (Vocabulary + Pronunciation) Our flagship endless-runner game transforms vocabulary practice into an action-driven experience. Instead of tapping buttons, players must speak French words out loud in real time—correct pronunciation triggers the jumping movement. It’s a fun mechanic, but it also solves a real industry gap: most apps test recognition, not actual pronunciation.
Whack-a-Mole! (Grammar) Grammar is traditionally one of the hardest parts of learning French. Our whac-a-mole–style game tackles this by presenting sentences with carefully crafted errors, such as the tense mismatches, agreement issues, missing articles, incorrect prepositions, and more. As each mole pops up holding the specific word or grammatical element in question, players must quickly tap the one containing the mistake. This fast-paced interaction trains intuition and reinforces pattern recognition, helping learners develop a natural feel for French grammar that usually takes months of traditional study to achieve.
Chat Learning (Conversation + Explanation) To complement the active gameplay, we developed a conversational learning module powered by OpenAI. With careful prompt engineering, we shaped the AI to act as a quick, friendly French tutor, someone who can guide learners through questions, explanations, and real conversation practice. It adapts to each user’s proficiency level, delivers clear contextual explanations, and provides a judgment-free space where learners can build confidence speaking and experimenting with the language.

Together, these modes create a rounded learning experience that touches vocabulary, grammar, and conversation, which are the essentials of true language proficiency, while keeping learners fully engaged through gameplay and immediate feedback.

How we built it

Duonosaur is built with a modern, scalable architecture designed for speed, reliability, and future growth.

On the frontend, we use React 19 with TypeScript, supported by Vite for rapid builds and a smooth developer workflow. Tailwind CSS and Radix UI give us a clean, unified design system that feels polished and professional from day one.

For speech recognition, we rely on the Web Speech API, processing audio directly in the browser. This gives us near-instant response times, which is essential for a fast-paced game with minimal latency, where a half-second delay would break the experience. It also keeps user audio private by avoiding unnecessary network transmission.

Our backend runs on FastAPI, chosen for its speed, strong typing, and clean integration with Python’s AI ecosystem. We used prompt engineering to generate vocabulary challenges and grammar exercises tailored to real learning needs. Each feature, whether for vocabulary, grammar, or chat, is isolated in its own router, making the system modular and easy to scale.

We also built robust error handling, CORS management, and fallback strategies to ensure the platform stays reliable even when AI calls take longer than expected or external services fluctuate. All of it was designed with extensibility in mind so new languages, game modes, or adaptive difficulty systems can be added without rethinking the foundation.

Challenges we ran into

Building Duonosaur wasn’t easy. One of the biggest technical challenges was achieving consistent, accurate, real-time speech recognition for French. The Web Speech API isn’t optimized for every accent or phonetic nuance, so we spent a lot of time tuning configuration, confidence thresholds, and event handling. We also had to engineer a robust state-management system to make sure players’ spoken inputs didn’t trigger duplicate moves or get lost in timing conflicts. That required custom hooks, debouncing strategies, and a deep understanding of how React processes asynchronous events.

On the AI side, generating “good” grammar mistakes turned out to be surprisingly challenging. We didn’t want random or nonsensical errors. We needed educational ones that actually teach something and have the users actually learn French. That required several iterations of prompt engineering design and a deeper dive into French grammar rules.

Finally, synchronizing AI-generated content with real-time gameplay pushed us to optimize rendering, reduce unnecessary component re-renders, and ensure the game consistently hits smooth frame rates. Balancing performance, UX, and AI latency was a recurring challenge that shaped much of our architectural approach.

Accomplishments that we're proud of

We’re proud that Duonosaur is more than just another educational app. It’s a fully interactive, polished platform that combines modern game design, real-time voice interaction, and AI-generated exercises in a single seamless experience.

We successfully built:

A voice-controlled gameplay system that requires actual pronunciation, not just tapping.
AI-assisted grammar exercises that teach real rules through carefully crafted errors.
A unified, professional user experience that rivals commercial platforms.
A modular backend that’s clean, scalable, and future-ready.
A complete product delivered under intense time constraints, thanks to strong prioritization and teamwork.

Most importantly, we proved our biggest hypothesis: when language learning feels like a game, people stay motivated longer, practice more consistently, and ultimately learn faster.

What we learned

Building Duonosaur taught us far more than just the technical skills required to ship a complex product—we gained insights into user behavior, learning psychology, and the realities of designing AI-powered educational tools. We learned that language learning becomes dramatically more effective when users are actively involved, receiving immediate feedback rather than passively consuming content. We discovered the importance of thoughtful prompt engineering, where even small adjustments can mean the difference between an exercise that feels meaningful and one that feels random. On the engineering side, we gained hands-on experience managing real-time systems, optimizing performance under constant user input, and balancing AI latency with smooth gameplay. Perhaps most importantly, we learned the value of building with empathy—understanding where learners struggle, what keeps them motivated, and how to create experiences that feel supportive rather than intimidating. These lessons will continue to guide us as we expand Duonosaur and refine it into a platform that truly elevates how people learn languages.

What's next for Duonosaur

Our roadmap is ambitious and focused on long-term impact.

We’re building:

A full analytics and progress-tracking system that learns from user behavior and adapts difficulty dynamically per user.
New game modes, including listening challenges, sentence-building activities, and cultural context games.
Cross-device user accounts so progress carries seamlessly across sessions.
Social and community features like leaderboards, friend challenges, and achievements.
Support for more languages—Spanish, German, Italian, and beyond.

On the technical side, we plan to develop native mobile apps, where voice recognition and performance can be even more precise. We’re also exploring offline experiences, smarter adaptive AI, certification prep content, and potential partnerships with schools and institutions. Our vision is to build the most engaging, effective language-learning platform in the world, one that makes speaking a new language feel exciting, intuitive, and genuinely fun.