Habitat

Inspiration

We were inspired by the joy of going down rabbit holes — when you start with one topic and keep diving deeper into related ones until you’re far from where you began. Wikipedia and YouTube make this easy, but chaotic. We wanted to turn this curiosity-driven process into something visual, structured, and interactive, so users can look back and see the paths they've taken.

What it does

Habitat is a knowledge exploration tool that generates an infinite graph of related topics.

Start with a root node
Click a child node to generate more related topics
Use Burrow to dive into a new layer, representing a rabbit hole journey
Explore previews from Wikipedia or YouTube by hovering or opening fullscreen
Watch as AI agents suggest new branches and help grow your knowledge map
It transforms curiosity into a living map of discovery.

How we built it

Frontend: Next.js (v15.5.3), React, and TypeScript for the graph and layered UI
- Node graph visualization with hover previews
- Layered view shows the “burrowing” journey
Backend: Node.js + Express with REST APIs
Agents: Mastra framework calls gemini to fetch content for a node
- Generate N related topics
- Spawn nodes with associated links (e.g. Wikipedia, YouTube)

Challenges we ran into

Gemini API errors: calls failed, breaking the agent flow
Graph performance: rendering many nodes caused lag, forcing us to balance speed vs fun animations
Navigation design: building a graph that’s infinite yet still usable required experimenting with both 2D canvases and 3D layered views
Time: co-op exploration, portal invites, and Chrome extension scraping had to be left as future features

Accomplishments that we're proud of

Built a full end-to-end pipeline from node creation → Gemini topic generation → new node visualization
Designed a Burrow mechanic that makes exploration feel like an actual rabbit hole journey
Created a foundation for AI + human co-exploration of knowledge spaces

What we learned

How to combine generative AI with graph data structures for topic expansion
The importance of UI tradeoffs between performance and immersion
How server-side agents can act as curiosity engines, automating topic exploration
New experience with Next.js (Turbopack), TypeScript everywhere, and AI API integration

What's next for Team 67

Collaboration: co-op exploration where two users can meet inside the same rabbit hole
Persistence: Supabase integration to save and reload rabbit holes
Cross-platform: Chrome extension to launch a rabbit hole from any webpage
Better grounding: combine Gemini with Wikipedia + YouTube scraping for higher-quality results
Portal invites: share a specific node with a friend to let them start their own rabbit hole