Inspiration

As developers, we often have heard about the term "try to make a rubber duck convinced about your idea first" to validate our knowledge. We applied this rule for every field,

What it does

You type any topic and an agentic pipeline (MiniMax, Exa, Retell AI) generates a knowledge graph, curates web resources into micro-courses, then tests your understanding through Socratic teach-back to an AI rubber duck — scoring accuracy, simplicity, structure, transfer, and metacognition.

How we built it

We used next js full stack + supabase + vercel(deployment) + minimax api for llm + minimax voice + retellAI api for real-time conversational ai agent using minimax voice.

For IDE we used cursor + opus 4.6 and 100$ usd worth of credits for 2 developers LOL

Challenges we ran into

optimizing graph interfaces in the frontend and making them as much customizable as possible (as less rigid as possible)

Accomplishments that we're proud of

The proudest accomplishment we've got so far is that we are truly passionate about the idea and we truly believe that it can be a great add-on for ANY edTech startup. It is as simple as a bottle of water on the desk but so brilliant.

We were able to create something outside of just a gpt wrapper for the first time for the whole team.

we were trying to set up a real-time conversational ai agent manually for 3 hours and have managed to make it but the latency was extremely high. Ultimately, we just added retell ai integration in one prompt.

What we learned

How easy it has become to build as big projects as we built in the last 15 hours. Through vibecoding, the field of software is truly being open to many new inventions and projects. We have also learnt to properly set Auth because we once got 400 auth requests in 5 minutes LOL

What's next for DuckLab

Built With

  • exa
  • mcp
  • minimax-llm-api
  • minimax-voice
  • next.js
  • supabase
  • ts
  • tsx
  • vercel
Share this project:

Updates