Epsilon Delta

Dashboard
Call your tutor
Custom Feedback
Genering manim videos

Inspiration and Philosophy

Our primary inspiration behind Epsilon-Delta was the youtube channel, 3blue1brown, who is adored by millions of college students for his visual and intuitive explanations of complicated mathematics topics. We believed that we could use the open-source Manim library and the Eleven Labs interface to take this a step further: an agentic platform that provides personalised videos, explanations, and guidance for university-level maths. We were also inspired by the Knowunity application and strived to replicate its intuitive dashboard and user interface.

What it does

Epsilon-Delta provides every user with their own custom agent, uniquely named after a famous mathematician. Users can solve problems or upload assignments on our interactive whiteboard to receive real-time guidance, step-by-step feedback, or full solutions when stuck. The platform also generates task-specific explanatory videos using Manim with voiceovers powered by ElevenLabs, tailored exactly to the user’s course and needs. Through voice (or text) interaction, the agent allows hyper-personalisation of learning, adapted to the user’s preferences, pacing, and learning needs, accounting for various learning styles and support for ADHD and dyslexia.

How we built it

We built Epsilon Delta using Typescript for the front-end and a cloudbase server and python for the backend. The manim generator was built using a custom library extracted from Manim documentation, including all relevant graphic tags for mathematical reasoning. Then our model parses this into manim code that is then rendered onto the front-end using the React animation library.

Challenges we ran into

One of our biggest challenges was ensuring the system accurately interprets user meta-data in real time, especially translating handwritten or spoken mathematics into concise and clear responses. We also faced challenges in integrating the Manim backend, especially with respect to creating our custom library since we found that a simple Claude wrapper was insufficient for this problem. Furthermore, integrating our various features together was a complicated task that involved untangling tricky merge conflicts.

What's next for Epsilon Delta

Because of our use of persistent state throughout the application we were also unable to deploy Epsilon-Delta onto serverless platforms like Vercel or Netlify. We intend to host the platform on our own custom VPS in the future. Furthermore, we would also love to add a new paradigm for tutors to directly interact with their tutees agents and automatically distribute practice papers, quizzes, and exam questions.