EduMotion AI

Inspiration

Every great teacher knows that a good animation beats a thousand words. But tools like Manim take weeks to learn, and most educators simply don't have that time. I wanted to remove that barrier entirely — so anyone could create a 3Blue1Brown-style video just by describing an idea in plain language.

What it does

EduMotion AI takes a text prompt (e.g. "explain bubble sort step by step") and returns a fully rendered MP4 animation with a natural voiceover — in under 2 minutes. It supports Arabic and English, with proper RTL layout for Arabic users.

How we built it

Google Gemini 2.0 Flash — generates valid Manim Python code from the prompt
Manim — the mathematical animation engine used by 3Blue1Brown
edge-tts — natural text-to-speech narration
FastAPI + Docker — production-ready backend deployed on Render
React + Vite — clean, responsive frontend deployed on Vercel

We also built a custom AI resilience engine with automatic API key rotation and model fallback — so if one Gemini model is overloaded, the system automatically switches to another without the user noticing.

A smart caching system using SHA-256 fingerprinting avoids re-rendering duplicate requests, saving time and compute.

Challenges we ran into

Getting Manim, LaTeX, Cairo, and FFmpeg to all work together inside a Docker container on a free server took many failed builds. We also hit async conflicts between FastAPI and the TTS engine, which we solved by switching to subprocess-based audio generation.

Accomplishments that we're proud of

Full end-to-end pipeline: text prompt → rendered MP4 in under 2 minutes
AI that generates valid, runnable Manim code — not just plausible code
Bilingual support (Arabic RTL + English LTR) with cookie-persisted settings
Resilient AI engine with automatic fallback across Gemini model variants
Production-deployed and fully accessible online

What we learned

The hardest part of an AI pipeline isn't the AI — it's everything around it. Containerization, environment isolation, caching, error recovery, and async handling are what separate a demo from a real product.

What's next for EduMotion AI

Community gallery to browse and remix generated videos
Export to PowerPoint slides alongside the video
Teacher dashboard for managing and sharing content with students
Support for longer videos and custom animation styles

Built With

docker
edge-tts
fastapi
ffmpeg
google-gemini-api
manim
python
react
render
vercel
vite

Submitted to

Created by

I built EduMotion AI end-to-end — from the AI pipeline to the frontend.

On the backend, I designed a resilient FastAPI server that integrates Google Gemini to generate Manim animation code and narration scripts in a single API call. I implemented a multi-key rotation system with exponential backoff to handle Gemini's free-tier quota limits without any downtime for the user.

I then wired up Manim (the mathematical animation library) as the rendering engine, added a pre-run code validator to catch errors before execution, and used Microsoft Edge TTS for neural Arabic/English voiceover generation. FFmpeg handles the final audio/video mixing.

On the frontend, I built a bilingual (Arabic RTL + English) React/Vite UI with smooth animations and a real-time progress tracker.

It was my first time combining Manim with a live API pipeline at this scale — getting the animation rendering to work reliably inside a web server environment was the biggest challenge, but I learned a tremendous amount about async processes, subprocess management, and AI prompt engineering.

Mohammed Salahuldeen

Updates

Mohammed Salahuldeen started this project — Apr 22, 2026 12:20 AM EDT

Leave feedback in the comments!

Log in or sign up for Devpost to join the conversation.