TutorQuest

AI-powered teaching assistant that generates personalized long division problems, validates answers, and creates custom Manim video explanations when students struggle.

Comment

Inspiration

Github: https://github.com/jv-code/TutorQuest

Math education often lacks personalized, patient tutoring for every student. We wanted to create an AI teaching assistant that:

  • Adapts to each student's skill level
  • Provides unlimited practice problems
  • Generates custom video explanations on-demand
  • Never gets frustrated, always encourages learning

Traditional tutoring is expensive and doesn't scale. With AI and automated video generation, we can provide every student with a personal math tutor.

What it does

MathingAI is an interactive long division practice platform that:

  • Smart Difficulty Scaling: Questions auto-adjust from simple (48 ÷ 6) to complex (4-digit ÷ 2-digit) divisions
  • Progressive Feedback Strategy:
    • Attempts 1-2: Generic encouragement
    • Attempt 3+: Detailed feedback + video offer button
  • On-Demand Video Generation: Students choose when they need visual explanations
  • Real-Time Chat: AI teaching assistant answers questions and provides hints

How we built it

Tech Stack

  • Frontend: React 19 + Vite with Tailwind CSS and shadcn/ui components
  • Backend: FastAPI (Python 3.13)
  • AI: Claude Haiku 4.5 (questions/validation) + Sonnet 4.5 (video generation)
  • Database: Supabase (PostgreSQL + Storage)
  • Video Generation: Daytona sandboxes executing Manim Python code

Architecture

┌─────────────┐     ┌──────────────┐     ┌─────────────┐
│   React UI  │────▶│  FastAPI     │────▶│   Claude    │
│  (Frontend) │     │  (Backend)   │     │   API       │
└─────────────┘     └──────────────┘     └─────────────┘
                           │
                           ├────▶ Supabase (Database + Storage)
                           │
                           └────▶ Daytona (Video Generation)

Video Generation Pipeline

  1. Claude Sonnet 4.5 generates Manim Python code tailored to the specific problem
  2. Daytona sandbox is created and configured
  3. Manim renders the explanation video
  4. Video is uploaded to Supabase Storage
  5. Frontend polls status every 2 seconds until complete
  6. Student receives custom video explanation

Prompt Engineering

We use Jinja2 templates for all AI prompts:

  • generate_question.j2: Difficulty-scaled question generation
  • validate_answer.j2: Smart answer validation with progressive feedback
  • generate_manim.j2: Manim code generation for educational videos

Challenges we ran into

Manim rendering took 20-40 seconds. Optimized by:

  • Using -ql (low quality) flag for faster rendering
  • Implementing polling mechanism instead of blocking requests
  • Showing real-time status updates to keep students engaged

Accomplishments that we're proud of

  • Seamless Daytona Integration: Successfully orchestrated Claude + Daytona to generate custom educational videos
  • Smart Difficulty Progression: Algorithm that adapts to student performance in real-time
  • Polished UX: Smooth polling, loading states, and error handling create a professional experience
  • Cost-Effective AI Usage: Strategic use of Haiku (fast/cheap) vs Sonnet (powerful) models
  • Production-Ready: Fully functional database schema, error handling, and video retention policies

What we learned

Technical Learnings

  • Daytona Sandbox Architecture: How to programmatically create, execute code in, and manage ephemeral development environments
  • AI Model Selection: When to use Haiku vs Sonnet based on task complexity and cost
  • Video Processing: Base64 encoding, storage, and streaming for generated content
  • Async Workflows: Building polling mechanisms for long-running tasks

Educational Insights

  • Progressive feedback is more effective than immediate detailed explanations
  • Students prefer choosing when to watch videos rather than forced viewing
  • Visual explanations (Manim) combined with text create powerful learning experiences

What's next for MathingAI

Immediate Roadmap

  • Multi-Topic Support: Expand beyond long division to fractions, decimals, algebra
  • Voice Explanations: Add text-to-speech narration to Manim videos
  • Progress Dashboard: Visualize student improvement over time
  • Gamification: Badges, streaks, and achievements for motivation

Long-Term Vision

  • Classroom Integration: Teacher dashboards to monitor multiple students
  • Custom Problem Sets: Let teachers upload their own problems
  • Peer Comparison: Anonymous benchmarking to encourage healthy competition
  • Mobile App: Native iOS/Android apps for on-the-go practice

Daytona-Specific Features

  • Multi-Language Support: Generate explanations in different programming languages
  • Interactive Notebooks: Create Jupyter notebooks with step-by-step solutions
  • 3D Visualizations: Use Daytona to render complex 3D math visualizations

Built With

Share this project:

Updates