Socratic

• 

  Start

• 

  you ask question and start a conversation

• 

  conversation

• 

  test you'r understanding

Inspiration

Most AI study tools just hand you the answer, which feels helpful but does not stick as well as figuring something out yourself. Real tutors ask questions instead of lecturing. I wanted to build something that does that by default, instead of relying on the user knowing to ask for it.

What it does

You tell it what you are trying to understand. Instead of explaining, it asks a short guiding question based on what you say, then keeps building on your answers until you are close. At any point you can hit "test me," write your own explanation in your own words, and it gives you an honest verdict on what you got right and what is still missing, not just generic praise.

How I built it

Next.js and Tailwind CSS on the frontend. One backend route handles two modes: normal Socratic questioning, and an evaluation mode that swaps in a different system prompt to grade the student's own explanation instead of asking another question.

Challenges I ran into

At one point the evaluation mode was returning completely empty replies. It turned out the model was running in a hidden reasoning mode by default, using its whole token budget thinking before it ever got to write the actual answer. I had to explicitly turn that off to get it to reply reliably.

What I learned

How to build one API route that switches behavior based on context instead of just having one fixed system prompt, and a real lesson in checking the raw API response instead of assuming an empty reply is a prompt writing problem.

What is next

Remembering which concepts a student has already worked through, and picking harder or easier starting questions based on how their last check went.

Built With

• deepseek-api
• next.js
• node.js
• react
• tailwind-css
• typescript