• 

  Architecture Diagram

What it does

🚀 Inspiration

The idea for this project came from a simple yet powerful question:
"What if anyone could create structured, high-quality educational content — just by typing a prompt?"

We saw how difficult and time-consuming it is for educators and learners to design curriculum, generate practice material, and create engaging visual content. Our goal: make AI a true co-pilot for education.

🛠️ What We Built

We created an AI-powered web platform that allows users to:

1. Enter a learning prompt (e.g., "Teach me supervised learning").
2. Automatically generate a main topic, subtopics, and for each subtopic:
   • Multiple-choice questions (MCQs)
   • (Upcoming) Structured reading content
3. Generate animated explanatory videos using Manim for selected topics.
4. Preview generated content via an interactive UI with:
   • Scrollable subtopic cards
   • Real-time selection feedback
   • Embedded video preview

The backend uses a multi-agent architecture to handle generation and rendering tasks modularly.

💡 What We Learned

• How to break down unstructured user input into structured learning modules using LLMs.
• Building a seamless pipeline between frontend selections and backend content generation.
• Designing intuitive UX for educational tools (balancing structure with flexibility).
• Generating programmatic animation using Manim, integrating AI and visuals effectively.
• Use of ADK for building multi-agent systems and workflows.
• Integrating Gemini-based LLMs effectively

🧩 Challenges We Faced

• Balancing AI creativity with curriculum structure: Ensuring content was pedagogically sound, age-appropriate, and coherent.
• MCQ generation quality: Validating that each question had exactly one correct answer and meaningful explanations.
• Manim integration: Creating readable and engaging video animations without overloading the screen.
• Scalability: Designing a backend that could scale content generation and video rendering reliably.
• Learning and integrating Agent Development Kit (ADK)

✅ Accomplishments that we're proud of

• Architected and deployed a modular AI system with agent-based design
• Enabled secure video rendering using isolated subprocesses
• Built a complete content-to-video pipeline from scratch
• Created an intuitive, interactive UI for end users
• Delivered a working prototype that bridges education and generative AI

🧱 How We Built It

• Frontend : Built using Next.js, deployed on Firebase
• Authentication : Powered by Firebase Auth, with user IDs stored in our backend
• Backend : Developed in Spring Boot, uses ADK for routing requests to agents
• Agents : Handle generation of topics, questions, and video content
• Video Rendering : ADK calls a Manim Running Agent, which sends code to a remote Manim server. The server runs the code in a subprocess, generates video, uploads to Google Cloud Platform (GCP)
• Video Playback : The UI retrieves and plays the video based on the returned job ID

🌟 What's Next?

• Add user profiles and learning progress tracking
• Add support for different languages and accessibility features
• Improve video rendering speed and introduce voice-over support
• Generate longer-form videos by combining multiple clips, with support for video editing

Built With

• adk
• cloudrun
• flask
• gcp
• gemini
• java
• manim
• nextjs
• postgresql
• python
• react
• springboot