Fun Learn

Inspiration Every year, millions of students fail exams not because they lack intelligence, but because passive learning methods like reading textbooks and memorizing answers are ineffective. We were inspired by the Nobel Prize-winning physicist Richard Feynman, who famously said, "If you can't explain it simply, you don't understand it well enough." We wanted to flip the traditional education model: instead of AI teaching students, the students should teach the AI.

What it does Fun Learn creates an active learning environment where Gemini acts as your student, not your teacher. The core Feynman Engine features Ritty, an AI persona of a curious 8-year-old who asks probing "Why?" questions, forcing users to deeply understand concepts. Other features include Misconception Cascade Tracing (MCT) which diagnoses the root cause of errors like a diagnostic surgeon, Time Travel Interviews with historical figures in their native languages, and a Debate Arena to build critical thinking skills.

How we built it The application uses a modern tech stack with React and TypeScript on the frontend and FastAPI with Python on the backend. The core intelligence is powered by Google Gemini (specifically Gemini 3 Pro Preview and Gemini 2.0 Flash) for reasoning and Imagen 3 for generating educational visual aids. We implemented a provider abstraction layer to manage AI interactions and used Google Cloud Text-to-Speech and Speech-to-Text for full voice accessibility.

Challenges we ran into Prompt engineering was our biggest hurdle. Tuning Gemini to genuinely act like a curious, slightly confused child—without hallucinating facts or becoming too helpful—required multiple iterations. Managing context across the Misconception Cascade Tracing to ensure the AI remembered specifically where the student went wrong while guiding them back was also complex. Implementing seamless switching between 9 languages while maintaining the unique personality of historical figures also posed significant architectural challenges.

Accomplishments that we're proud of We are most proud of the native multi-language support. With a single click, the entire application—including historical personas like Marie Curie—switches to speaking native Hindi, Bengali, or Japanese, adopting the correct cultural nuances. We are also proud of the Learn from Anything feature, which successfully identifies educational angles from random photos, turning everyday objects into learning opportunities.

What we learned We learned that AI is surprisingly effective at role-playing ignorance. By instructing Gemini to feign confusion, we unlocked a powerful pedagogical tool that drives human engagement far better than a standard tutor chatbot. We also deepened our understanding of multimodal architecture, effectively combining vision, voice, and text generation into a cohesive, low-latency experience.

What's next for Fun Learn We plan to expand the Feynman Engine to support multiplayer classrooms where students can collaborate to teach the AI together. We also aim to implement AR features where the Learn from Anything mode can overlay physics vectors or historical context directly onto the camera feed in real-time.

Built With

• apis
• cloud-services
• databases
• fastapi
• frameworks
• gemini-3
• give-me-few-keywords-separated-by-semi-colon-built-with-what-languages
• google-cloud
• interactive-learning
• platforms
• python
• react
• tailwindcss
• typescript
• vite