Software · Multi-agent · 36-hour hackathon

Curiosity Catalyst

an AI tutor, Plato the cat, that learns how you learn

LLMs at your fingertips can quietly erode critical thinking. We wanted the opposite. In 36 hours we built Curiosity Catalyst, starring Plato, a very curious, very alive cat that lives on your screen.

What it does

Plato floats on your desktop, watches how you work, reads the content on your screen, and infers your level of understanding. When the moment is right he opens a side panel with three lenses of learning: conceptual, to strengthen foundations; application, to iterate on problems; and extension, to push your understanding further.

Using Bayesian Knowledge Tracing, Plato builds a confidence-weighted map of what you actually know and deploys the right agent, switching between them dynamically and completely unprompted. When it helps, he generates a custom visualization to layer abstractions, or rings you with a live speech-to-speech phone call paced to your understanding to walk you through a concept.

How we built it

The live screen feed runs through a two-pronged approach: Gemini’s vision model and Zoom’s screenshare render analyze the stream of content and your activity. Bayesian Knowledge Tracing decides when to intervene. Three learning agents, one per lens, are built on fetch.ai’s agent tooling, hosted on AgentVerse, and findable on ASI:One, with an orchestrator agent handling seamless handoffs between them. Visualizations are generated on the fly, and OpenAI’s speech-to-speech API powers the “incoming call.” Plato himself is our own art, brought to life with help from Sora.