LearnFlows

Visual Learning Powered by Perplexity and Manim

Inspired by 3Blue1Brown — the popular YouTube creator who uses Manim animations to make complex math intuitive and beautiful.

I built LearnFlows, a generative Manim-based app that simplifies advanced concepts through clear, interactive visual storytelling — all powered by the Perplexity Sonar API.

🚀 How It Works

User Intent Refinement — Clarify the learner’s goal.
Content Analysis — Identify topic depth and key elements.
Explanation Generation — Produce simple, accurate explanations.
VideoFlow Construction — Break explanations into storyboard steps.
Director Module — Convert storyboard into Manim scripts.
Code Improvement + RAG — Refine scripts with Retrieval-Augmented Generation using Manim docs.
Execution Check
✅ Runs successfully? Render and display video.
❌ Errors? Send to ErrorSolverAgent for fixes, retry until success.

🛠️ Tech Stack Highlights

LLM Backbone

Perplexity Sonar API – For deep topic analysis
Gemini 2.0 Flash – Fast reasoning and intent handling
DeepSeek Coder V1 – Specialized in Manim scripting

Frontend

Next.js + Supabase

Backend

FastAPI

Rendering

Manim + FFmpeg

Vector Search & RAG

Qdrant – Retrieval-augmented generation using Manim documentation

LLM Integration

Vercel AI SDK – Seamless frontend-to-LLM interaction

Error Handling

ErrorSolverAgent – Iterative debugging and code repair

📈 Future Plans

🎙️ Add TTS voice-over narration
🌍 Expand into physics, computer science, and broader STEM topics
🧩 Enable interactive clickable videos
📄 Support user LaTeX/code snippets for auto-visualization
🌐 Launch a social learning platform
🎬 Modular scene-wise video generation for better structure
🚢 Full public launch with personalized learning paths

🙏 Thank You

Built for the Perplexity Hackathon with love for AI, education, and storytelling.
Special thanks to Perplexity Sonar, Manim, and the amazing open-source community.

Let’s make visual learning smarter, clearer, and more creative.