3Brown1Blue

Generate a high-quality, detailed animated video to help you understand any concept in seconds. This project interfaces custom LLM agents with the Manim library to bring this vision to users.

Comment

Inspiration

We had the vision to create an AI-powered video generation "teacher", which meets the quality of educational videos created by the YouTuber 3Blue1Brown. For us, and millions around the world, 3Blue1Brown was able to make the most difficult to visualize topics much more intuitive by creating a medium where impossible meets possible: animation. His videos are made in a digestible format with mindblowing visualizations, we hope to emulate some of that in an automated process, such that users can learn about whatever they would like in a similar way.

What it does

3Brown1Blue (Grant Sanderson) creates high-quality AI generated videos explaining any concept - it's the perfect personal tutor. While generating the code, feel free to speak to your personal Empathetic Voice Interface (EVI) to get your mind of the stress of learning new concepts. If that's not your style, we also have the Dinosaur game from Chrome to keep you busy! We're here to make the uncomfortable gap between being a novice and mastering the concept nonexistent in the most seamless way possible.

How we built it

  • Python + FastAPI for API routing and middleware
  • Hume AI API for expression measurement and Empathic Voice Interface
  • LMST API for Text to speech, and to emulate Grant's voice
  • OpenAI, AWS Bedrock, Anthropic, Groq for pluggable LLM inference
  • React, Next + TypeScript for frontend development
  • Tailwind, ShadCN for UI
  • Manin for video generation and animation

Challenges we ran into

Throughout the course of our development of 3Brown1Blue, we had major design decisions fall through due to the acquisition of new information. Our initial vision was a Google Meet style app where you would chat with your virtual teacher and see explanations as a video as you typed questions and queries in the chat section. We believed that Hume AI exposed a pure speech to text model, but due to the added feature of the EVI, we had to figure out how to incorporate her into our application. This led to on the fly, high pressure scenario where with little time left, we had to put together a features that recently lost purpose. What came together was magical. We created a multi-purpose app where the barriers to education break down.

Accomplishments that we're proud of

  • Getting an MVP out really quickly
  • Getting an AI to generate and run code based on prompts
  • Learning about implementing multiprocessing and websockets

What we learned

  • Spend time thinking before you build
  • LMNT
  • Hume AI
  • Groq
  • AWS Bedrock

What's next for 3Brown1Blue

We believe there is huge potential for the development of visual learning with AI. We hope to continue developing our project to generate the best videos as fast as possible.

Built With

Share this project:

Updates