About the Project
Inspiration
Advanced mathematics is often presented in its symbolic form before its underlying intuition is made clear. As a result, many learners can manipulate equations without forming a concrete understanding of what they represent.
At the same time, mathematics is becoming increasingly central in the age of AI — yet intuitive access to it remains limited. We wanted to build a system that makes mathematical structure visible, interactive, and learnable through exploration.
What it does
Axiomatic transforms a single prompt into a complete, interactive learning experience.
Given a concept such as eigenvalues, integrals, or Fourier transforms, the system automatically generates:
Structured lesson plans — decomposing topics into core concepts and dependencies Animations with narration — automatically generated visuals that make structure visible Interactive environments — allowing users to manipulate parameters and observe behavior in real time Adaptive quizzes and follow-ups — evaluating understanding and guiding what comes next
Everything is grounded in dialogue with the user, allowing the system to refine explanations and adapt the learning process dynamically.
How we built it
Axiomatic is designed as a multi-agent system coordinated through shared state. Using langraph and via coordination of over 5 agents, we develop a pedagogy, quizzes based on time stamps in a video, interactive elements directly related to the ideas presented in a video, manim code via an agent that renders on a per scene basis, and audio using the eleven labs api based on a script generated by a sub agent as well.
A lesson planning agent decomposes the topic. A visualization agent generates animations using Manim. An assessment agent generates and evaluates quizzes.
These agents are orchestrated using LangGraph, which enables persistent context across interactions, dynamic routing between components, and adaptive learning loops based on user feedback.
A lightweight knowledge graph tracks conceptual relationships and user progress, guiding learners through a coherent path.
Challenges we ran into
Synchronizing animation, narration, and interaction into a single coherent lesson Ensuring explanations are both mathematically correct and intuitive Balancing real-time generation with high-quality visual output Designing agent coordination that feels unified rather than fragmented
Accomplishments that we’re proud of
Building a system that generates complete multimodal lessons from a single prompt Integrating animation, interaction, and assessment into a unified experience Designing a coherent multi-agent architecture that adapts in real time Making advanced mathematical ideas intuitive without simplifying them away
What we learned
Agentic systems enable structured, adaptive workflows beyond single LLM calls Visual and interactive representations significantly improve conceptual understanding Shared state is essential for coordinating multiple components System cohesion matters more than adding additional features
What’s next for Axiomatic
Expanding coverage across broader areas of mathematics and related fields Improving real-time performance for animation and interaction generation Developing richer user models to better personalize learning paths Scaling the platform to support widespread access to intuitive math education
Log in or sign up for Devpost to join the conversation.