OpenCurriculum

Inspiration

Learning today is fragmented. Students jump between videos, textbooks, and explanations without a clear sense of what to learn next or how concepts connect. Even with powerful AI tools, learning often feels passive—ask a question, get an answer, move on.

We wanted to reimagine learning as an interactive, guided experience: one where students can explore ideas visually, navigate concepts naturally by voice, and learn through focused micro-lessons instead of overwhelming lectures. Our goal was to combine conversation, structure, and visualization into a single learning flow.

What it does

OpenCurriculum is a voice-driven learning platform that pairs a live AI tutor with a dynamic knowledge graph.

Students join a live session with an AI tutor in a video call
A 3D knowledge graph shows how concepts relate to one another
Students navigate topics by clicking nodes or using natural voice commands
Selecting a topic triggers a single, focused micro-lesson
The system generates a visualization best suited for that lesson
Students can return to the graph and continue exploring seamlessly

The result is learning that feels guided, exploratory, and interactive rather than static or linear.

How we built it

Frontend: A clean Next.js interface with a split-screen layout—live video on the left and the learning panel on the right—styled with Tailwind and shadcn/ui
Realtime interaction: Live video rooms power real-time communication between the student and the AI tutor
AI agent: A live AI tutor interprets natural language commands and controls navigation through the learning experience
Speech: Speech-to-text (send-on-silence) and text-to-speech enable natural voice interaction
Knowledge graph: A 3D force-directed graph visualizes relationships between concepts and updates as the learner progresses
Backend: A FastAPI service manages session state, graph data, and lesson selection
Visualizations: Depending on the lesson, the system generates an appropriate visualization using interactive 3D scenes, mathematical animations, or AI-generated diagrams

Each lesson is intentionally scoped to a single concept to keep learning focused and digestible.

Challenges we ran into

Designing natural voice navigation while keeping the system deterministic and reliable
Managing real-time audio, video, and data communication without disrupting the learning flow
Balancing fast UI responsiveness with slower AI-driven generation tasks
Keeping the interface clean and intuitive despite complex underlying systems
Coordinating multiple real-time components under hackathon time constraints

Accomplishments that we're proud of

Building a fully voice-navigable learning experience
Integrating a real AI tutor into a live video environment
Creating a dynamic knowledge graph that adapts as learning progresses
Delivering a polished, non-gimmicky interface under tight deadlines
Designing a system that emphasizes conceptual connections rather than isolated answers

What we learned

We learned that effective learning tools are not just about stronger models—they are about interaction design. Small decisions around flow, timing, and visualization can dramatically change how understandable and engaging a lesson feels. We also gained hands-on experience integrating real-time systems, AI agents, and frontend UX into a cohesive product.