TutorTalk

Inspiration

TutorTalk started from the frustration of grinding vocabulary apps that never actually make people comfortable speaking out loud. Instead of flashcards, the goal was to let learners practice real conversations on topics they already love, with an AI tutor that is always available and never judges mistakes.

What it does

TutorTalk is a web app where learners have live, free‑flowing conversations with an AI tutor that adapts to their level and interests. After each session, they get instant feedback on accuracy, vocabulary, and fluency, along with gentle corrections and suggested alternative phrases.

How we built it

The MVP is built with Next.js for the frontend/app router, Supabase for authentication and database, and Flowglad for metered AI token billing. Conversations are powered by OpenAI and Elevenlabs and streamed to the client, while Supabase stores user profiles and session history and Flowglad tracks “tokens used” per user so usage can map cleanly into pricing plans.

Challenges we ran into

Getting Supabase auth and profile creation to work reliably required careful trigger design; early on, a missing profiles table and failing trigger caused “database error creating new user” on sign‑up. Another big challenge was designing token-based billing so that Flowglad’s usage meters match our internal token counters without double‑counting or breaking the free tier experience.

Accomplishments that we're proud of

The team shipped a working, end‑to‑end AI tutor in hackathon time: new users can sign up, start a conversation, see tokens consumed, and understand exactly what they’re being billed for. It also sets up a clear business model with free, subscription, and metered “top‑up” style usage on top of a stack that’s simple enough for a solo indie dev to maintain.

What we learned

Designing AI products around tokens forces thinking about unit economics from day one, not as an afterthought. The hackathon also reinforced how much leverage modern tools like Supabase, Next.js, and developer‑friendly billing layers give: most time went into UX and pedagogy instead of boilerplate infrastructure.

What’s next for TutorTalk

Next up is adding richer speech features: real‑time pronunciation scoring, voice input, and speaking drills so it feels more like a live tutor than a chat app. On the business side, the plan is to refine plans and usage caps, launch a public beta, and experiment with niche communities (e.g., tech workers improving English for interviews or presentations) to find strong product–market fit.

Built With

• elevenlabs
• flowglad
• next
• react
• supabase