education
Adaptive Socratic Tutor
Guided reasoning that adapts to your concept-level weaknesses.
educationSocraticadaptiveconcept-trackingGemini
By
Aayush Nair
Semester
Spring 2026
Problem
Students using LLMs for studying often receive direct answers, which reduces conceptual understanding and long-term retention. There is no system that enforces guided reasoning while tracking concept-level weaknesses and adapting difficulty dynamically.
Solution
An adaptive Socratic tutoring platform where users enter a topic or upload notes. The system extracts structured concepts, generates tagged Socratic questions at configurable difficulty levels, evaluates reasoning quality, and dynamically adjusts future questions based on detected misconceptions. A dashboard visualizes stats, weaknesses, and learning progression.
User flow
- Enter a topic or upload notes
- System extracts structured concepts
- Receive tagged Socratic questions at the chosen difficulty level
- Reasoning quality is evaluated; misconceptions are detected
- Future questions auto-adjust based on detected weaknesses
- Dashboard tracks stats, weaknesses, and progression
LLM components
- Concept extraction — from unstructured text
- Controlled Socratic question generation — role-locked prompting
- Adaptive difficulty prompting — based on performance history
Tools
- Vibe coding: Claude Code, Cursor, Figma
- LLM: Gemini API
- Backend: FastAPI
- Frontend: React
- Database: Supabase
- PDF parsing: PyMuPDF