ManimAI

Math notation is everywhere. Real understanding is rare. Manim·ai transforms any math excerpt into a 3Blue1Brown-style explanation - turning static symbols into intuition you can actually see.

Comment 1

Inspiration

As a math major, I would love to see more visual understanding and explanation of concepts, as I feel that it is a much better way to learn. With ManimAI, I'm able to take concepts that are not readily available online and turn them into a visual experience to enhance my learning.

What it does

ManimAI takes a screenshot of a slide deck or even an excerpt from a textbook and basically gives you a three blue one brown style animated video with external resources, further clarification options as well as full code transparency for you to play around with as well

How we built it

We built it primarily using Claude code and with the extensive use of four main MCP servers:

  • Playwright
  • Context 7
  • GitHub
  • Tavily
  • Claude Design

Challenges we ran into

One of the challenges was figuring out what prompt to give the LLM on the backend for the further clarification one. We wanted to ensure that it is not very similar to the initial generation but also is more condensed and captures a fragment of the initial video generation. Figuring out the prompt to find something distinct yet within the same scope of the initial generation was something that took a lot of trial and error. After using Claude Design as well, we also found that there were other computational and deep backend issues that made it harder to deploy this on Vercel as Vercel mainly uses its own Next.js language, plus its own structure, which is slightly different to the Claude Design programming and structure

Accomplishments that we're proud of

We're really proud of the idea and how this idea can be applied to students anywhere in the world with access to a computer. We're also very glad that we were able to get the functionality of further clarification and a confidence score within the short time interval that we were given of one day.

What we learned

We truly were able to utilize the full capabilities of Claude Code. We were able to understand and use sub-agents as well as skills, and we were able to integrate those sub-agents and skills with our four MCP servers. We also, for the first time, used Claude Design.

What's next for ManimAI

Some of our limitations included the lack of optical character recognition plus compute power. I think we would like to explore cost-effective and feasible ways to scale our rendering so that it is quicker and a way that we can account for a large number of video generations. In terms of OCR, maybe partnering with companies and reaching out to firms that specialize in OCR technologies.

Built With

Share this project:

Updates