ADK Study Planner 🎓

Your Personal AI Study Companion for Personalized Learning

✨ Live Demo: https://academy.pinkcodequeen.com

Track: Agents for Good (Education)

📖 Overview

ADK Study Planner is an intelligent multi-agent system designed to help students organize their learning, discover high-quality resources, and understand complex academic concepts. By leveraging the power of Google's Agent Development Kit (ADK) and Gemini models, it creates a personalized learning experience that adapts to the student's topic, grade level, and language preferences.

🚩 Problem Statement

Students often face three main challenges:

1. Overwhelm: Breaking down complex topics into manageable study sessions is difficult.
2. Resource Discovery: Finding reliable educational videos and academic papers takes time.
3. Accessibility: Academic language can be dense and hard to understand, especially for non-native speakers.

💡 Solution

The ADK Study Planner solves these problems with a coordinated team of AI agents:

1. Planner Agent: Generates a structured 5-day study schedule with clear learning objectives.
2. Researcher Agent: Curates relevant YouTube videos and finds the latest academic papers on the topic.
3. Academic Agent: Simplifies complex abstracts and translates explanations into the student's native language (e.g., Persian).

🤖 Why Agents?

A simple script or chatbot wouldn't suffice for this application. Agents are the ideal solution because:

• Sequential Reasoning: The output of the planning phase directly informs the research parameters, requiring a system that can maintain context and pass state between distinct "experts."
• Autonomous Tool Use: The Researcher agent must dynamically decide what to search for on Google or Arxiv based on the specific nuances of the generated plan, rather than following a hard-coded search pattern.
• Adaptability: Agents can adjust their complexity and tone (e.g., simplifying quantum physics for a 10th grader vs. a university student) in a way that static code cannot.

🏗️ Architecture

The system uses a Sequential Multi-Agent Architecture orchestrated by a root coordinator agent.

graph TD
    User[User Query] --> Coordinator[Coordinator Agent]
    Coordinator -->|Step 1| Planner[Planner Agent]
    Coordinator -->|Step 2| Researcher[Researcher Agent]
    Coordinator -->|Step 3| Academic[Academic Agent]
    
    Planner -->|Generates| Plan[5-Day Schedule]
    Researcher -->|Finds| Resources[Videos & Papers]
    Academic -->|Simplifies & Translates| Content[Accessible Content]
    
    Plan --> Final[Final Response]
    Resources --> Final
    Content --> Final

Loading

Agents & Tools

• Planner Agent: Uses gemini-2.5-flash-lite to structure learning paths.
• Researcher Agent: Equipped with:
  • google_search: To find educational videos.
  • search_arxiv: A custom tool to fetch relevant academic papers.
• Academic Agent: Specialized in simplification and translation.

🔑 Key Features (Capstone Concepts)

This project demonstrates the following key concepts from the AI Agents Intensive:

1. Multi-Agent System: A sequential workflow where agents pass information and build upon each other's work.
2. Tools:
   • Built-in: google_search for video discovery.
   • Custom: search_arxiv tool using the Arxiv API for academic research.
3. Sessions & Memory: Uses InMemorySessionService to maintain conversation context and user preferences across the workflow.

✨ Web Interface Features

The modern web application includes several user-friendly features:

Feature	Description
🎤 Voice Input	Speech-to-text transcription using Web Speech API for hands-free query input
🔗 Smart Resource Buttons	Intelligent contextual buttons that auto-extract YouTube videos and arXiv papers from responses
📋 One-Click Copy	Easily copy the entire response with a single click
🌙 Dark/Light Theme	Toggle between dark and light modes for comfortable reading
💬 Modern Chat UI	Clean, Gemini-inspired interface with professional message styling
📎 File Attachments	Support for uploading documents (PDF, DOC, TXT)

🚀 Setup & Usage

Prerequisites

• Python 3.10+
• Node.js & npm (for the web interface)
• Google Cloud Project with Vertex AI API enabled (or AI Studio API Key)

1. Core Agent Setup

1. Clone the repository:

   git clone https://github.com/snowholt/Study-Planner.git
   cd Study-Planner/adk-study-planner

2. Install Python dependencies:

   pip install -r requirements.txt

3. Configure Environment: Create a .env file in the adk-study-planner directory:

   GOOGLE_API_KEY=your_api_key_here

4. Run the CLI Agent:

   python -m src.main

2. Web Application Setup (Optional)

To run the full web interface with chat history and user management:

Step A: Start the ADK Server

# In adk-study-planner directory
adk run src/main.py:root_agent --port 8081

Step B: Start the Backend API

cd webapp/backend
pip install -r requirements.txt
uvicorn main:app --reload --port 8000

Step C: Start the Frontend

cd webapp/frontend
npm install
npm run dev

Access the web app at http://localhost:5173.

📄 Example Output

Query: "Topic: Quantum Physics, Grade: 10, Language: Persian"

Result:

• Day 1-5 Plan: Introduction to particles, wave-particle duality, etc.
• Resources: Links to "Quantum Physics for Babies" video, Arxiv paper on Quantum Education.
• Simplified Content: "Quantum entanglement is like having two magic dice..." translated to Persian.

🔮 Future Improvements

• Calendar Integration: Automatically add the generated 5-day plan to the user's Google Calendar.
• Quiz Generation: Add a "Teacher Agent" to generate daily quizzes based on the study material.
• Voice Interface: Enable voice-to-text for queries. ✅ Completed!
• Text-to-Speech: Add audio playback for reading out the simplified summaries.
• Multi-Modal Inputs: Allow users to upload a textbook PDF for the agents to analyze and plan around.
• Progress Tracking: Track learning progress and provide personalized recommendations.

🏆 Submission Details

This project is submitted for the Agents Intensive Capstone Project.

• Track: Agents for Good
• Team: Individual

👤 Author

Nariman Jafari

• 🌐 Live Demo: academy.pinkcodequeen.com
• 📧 Email: hello@pinkcodequeen.com
• 👔 LinkedIn: linkedin.com/in/narimanjafari
• 🐙 GitHub: @snowholt
• 🏅 Kaggle: @snowholt