• 

  logo

MindLoom: Transforming Learning Through AI-Powered Visualization

Inspiration

MindLoom was born from a frustration with static, one-size-fits-all learning tools. Watching visual learners struggle with dense textbooks sparked the idea: What if complex concepts could come alive as personalized animations and quizzes? Inspiration came from:

• Manim (used by 3Blue1Brown for math visualization).
  
• Duolingo’s adaptive learning and gamification.
  
• The potential of generative AI to democratize education.
  

What It Does

MindLoom is an AI-powered chatbot that:

1. Visualizes Text: Generates 30-60 second animations (e.g., linked lists, sorting algorithms using Manim.
   
2. Creates Interactive Quizzes: Builds editable 2D arrays, multiple-choice questions, and live feedback.
   
3. Adapts to Users: Adjusts speed/difficulty based on behavior (e.g., "🐢 Slow Learner Mode").
   
4. Supports Multilingual Learning: Answers and teaches in Urdu/English, with plans for more languages.
   
5. Parses Uploads: Extracts answers from PDFs/images and explains concepts from user-provided docs.
   

How We Built It

Tech Stack

• Backend:
  
  • Python + LangChain (AI workflows), Gemini (content generation) and Langgraph (AI Agents)
• Frontend:
  
  • Next.js.
• Animation Engine:
  
  • Manim with Azure pre-rendering to reduce latency.
    

Challenges We Ran Into

1. Real-Time Rendering:
   
   • Problem: Manim animations took 5+ minutes to render.
     
   • Fix: Pre-rendered common topics + optimized video compression.
     
2. Urdu Technical Terms:
   
   • Problem: No standardized translations for CS concepts like "hash table."
     
   • Fix: Partnered with linguists to build a glossary.
     
3. Malicious Uploads:
   
   • Problem: Early file uploads crashed the system.
     
   • Fix: Added ClamAV scans and 5MB file limits.
     
4. UI Overload:
   
   • Problem: Users found too many buttons confusing.
     
   • Fix: Simplified menus and hid advanced features.
     

What We Learned

• Technical:
  
  • Pre-rendering is key for real-time animations.
    
  • AI-generated quizzes need human validation for accuracy.
    
• User-Centric Design:
  
  • Simplicity > feature overload.
    
  • A student’s complaint about Urdu translations led to massive improvements.
    
• Adaptability:
  
  • AI can’t replace human nuance—fine-tuning models for empathy (e.g., detecting frustration) is critical.
    

What's Next for MindLoom

1. Expand Languages: Add Hindi, Arabic, and Spanish support.
   
2. AR/VR Integration: Let users "walk through" data structures in 3D.
   
3. Collaborative Learning: Group animation projects and peer quizzes.
   
4. AI Tutor Avatars: Customizable mentors that guide users via voice/video.
   
5. Classroom Integration: Partner with schools to replace outdated CS textbooks.
   

Built With

• aiagents
• fastapi
• gemini
• langchain
• nextjs