Converso

🚀 Inspiration

• Online learning often feels passive and one-sided.
• I wanted something more like a real conversation with a tutor.
• Goal: use AI + voice to make learning personal, dynamic, and engaging.

🎙️ What it does

• AI-powered LMS where you learn through voice conversations.
• Build your own AI Companions (mentors) for any subject.

• Customize:

  • Name 🏷️
  • Subject / Topic 📚
  • Teaching style (formal / casual) 🎩
  • Voice 🎤

• Launch sessions, ask questions, and get spoken answers in real-time.

• Feels like one-on-one tutoring without the awkwardness.

🛠️ How we built it

• Frontend: Next.js 15 (App Router) + TypeScript + Tailwind CSS + Radix UI.
• Backend: Supabase (PostgreSQL) + Next.js Server Actions.
• Auth: Clerk (secure email/Google login).

• AI & Voice:

  • Google Gemini-2.5-flash-lite (conversational AI).
  • Web Speech API (speech recognition + synthesis).

• All integrated into a single, seamless full-stack app.

⚡ Challenges

• Syncing speech recognition + synthesis (don’t talk over the user!).
• Inconsistent browser support for the Web Speech API.
• Free-tier Supabase idling issue → solved with GitHub Actions ping.

🏆 Accomplishments

• Built a complete end-to-end voice-first platform.
• Smooth, real-time conversational flow 👏.
• Companion Builder: highly customizable AI tutors.
• Clever automation fix for database inactivity.

📚 What we learned

• Deep dive into Next.js App Router + Server Actions.
• Hands-on with Supabase, Clerk, Gemini integration.
• Design principles for voice-first UIs: timing, state, feedback.

🔮 What’s next

• 🌍 Multi-language support.
• 📊 Smarter feedback on learning progress.
• 👥 Multiplayer / group study sessions.
• 🧠 Improved conversational memory for long sessions.

Built With

• clerk
• gemini
• nextjs
• supabase
• typescript