Inspiration:
In a world where personalized learning is critical but quality human tutors are in short supply, we were inspired to create a solution that blends the empathetic power of human teaching with the scalability of AI. We imagined a platform where human tutors can train AI clones of themselves—allowing them to teach 24/7, support more students, and reach learners anywhere in the world. That vision became HumAI Tutor.
What it does:
HumAI Tutor is a next-gen hybrid tutoring app that enables:
- Human tutors to train AI replicas by uploading teaching content and recorded sessions.
- Students to have interactive AI-powered learning conversations—via text, voice, or real-time video.
- Live human tutor join-ins when the AI detects confusion or when students request personal help.
- A backend that tracks student progress, provides feedback, and adapts lessons based on prior learning history.
How we built id
Frontend: Built with Bolt, allowing rapid development of responsive and interactive UI components. Backend: Powered by Supabase for real-time database syncing, authentication, and user session tracking. Video Chat: Enabled by Agora API for seamless and low-latency real-time video communication between tutors and students. AI-Powered Conversations: Tavus powers our AI conversation cloning system—letting human tutors record sessions and turn them into interactive AI agents. Google Gemini serves as the core large language model (LLM), enabling natural, intelligent, and context-aware tutoring interactions.
Challenges we ran into
Multimodal sync: Integrating real-time video (Agora) with AI feedback (Gemini + Tavus) while maintaining a consistent user experience. AI style cloning: Capturing a human tutor’s tone and teaching style accurately in Tavus was more complex than expected.
Accomplishments that we're proud of
Successfully integrated five major technologies (Bolt, Supabase, Agora, Tavus, Gemini) into a seamless working MVP. Built a working trainable AI tutor clone system using human video input and Gemini-generated responses. Enabled real-time video tutoring fallback where AI gracefully escalates to human tutors when needed.
What we learned
How to orchestrate multimodal experiences (text, voice, video) in real-time applications. The importance of embedding human context into AI—style, pacing, and emotion matter in learning. Leveraging LLMs like Gemini effectively requires good retrieval and prompt strategy design.
What's next for HumAITutor
Integrate Payment Processing in the application Building an effective RAG pipeline that balances Gemini's creativity with fact-grounded tutoring.
Built With
- agora
- bolt
- gemini
- supabase
- tavus
Log in or sign up for Devpost to join the conversation.