FableLab

• 

  Homepage

• 

  Playground Mode

• 

  Lesson Page - Story

• 

  Lesson Page - Experiment

Project URL

• https://fable-lab.vercel.app/

Public Code URL

• https://github.com/emrsyah/FableLab
• https://github.com/emrsyah/FableLab-ADK

Gemini Integration

Fable Lab leverages the cutting-edge capabilities of Gemini 2.5 Flash (Native Audio) and Gemini 3 Flash via Agent Development Kit to transform passive learning into an interactive, voice-driven creative studio. The application is built on two core agentic pillars:

1. Real-Time Voice Facilitator (Live API BiDi Streaming in Agent Development Kit)

Powered by gemini-2.5-flash-native-audio-preview, our Playground Agent enables seamless, low-latency voice conversations. Students interact naturally with an AI "lab partner" that can hear their questions and instantly write or modify p5.js code to visualize concepts in real-time. This creating a "coding by conversation" experience that lowers the barrier to entry for STEM subjects.

2. Advanced Multi-Agent Pipeline (Agent Development Kit)

Our content generation engine uses a sophisticated multi-agent architecture to build complete lessons:

• Orchestration & Research: Specialized agents (Librarian, Plotter) use Gemini 2.5 Flash to structure lesson plans and research p5.js documentation via tool use.
• Code Generation: The P5MasterGenerator utilizes Gemini 3 Flash Preview (gemini-3-flash-preview) for its superior reasoning and coding capabilities, converting research notes into complex, bug-free interactive experiments.

Why Gemini? The integration relies on Gemini's multimodal capabilities (accepting audio, text, and images of the user's workspace) and Bidi-Streaming to maintain the flow of conversation. By combining structured Pydantic outputs with creative reasoning, Gemini acts not just as a chatbot, but as a full-stack educational engine that designs, codes, and teaches simultaneously.

Built With

• adk
• gemini3
• livestreamapi
• nanobanana
• next
• python
• texttovoice