Google ADK 多智能体实战：个人助理与 7 智能体学习计划器

A complete multi-agent AI system built with Google Agent Development Kit (ADK), featuring a personal assistant and an intelligent multi-agent study planner.

📖 Overview

This repository is a complete implementation of the Google ADK Workshop — a hands-on exploration of building intelligent AI agents with Google's open-source Agent Development Kit.

The project covers two distinct agent architectures:

Task 1 demonstrates a single-agent system with custom Python tools, capable of reasoning across tools to answer complex multi-part questions.
Task 2 demonstrates a multi-agent orchestration workflow — with sequential pipelines, parallel execution, and iterative review loops — to produce a high-quality, personalized 10-day exam study plan.

Note on LLM Provider: Google Gemini's free tier has aggressive quota limits on newer models. This project is configured to use the Groq API (specifically qwen/qwen3.8-27b) via the litellm extension — providing a completely free, high-speed, and reliable alternative without any billing setup.

🗂️ Project Structure

Google-ADK-AI-Agents/
│
├── Task1_Personal_Assistant/       # Single-agent personal assistant
│   ├── agent.py                    # Agent definition + custom tools
│   └── __init__.py
│
├── Task2_SmartStudy_Planner/       # Multi-agent study planner workflow
│   ├── agent.py                    # Full 7-agent workflow definition
│   └── __init__.py
│
├── docs/
│   └── google_adk_workshop_task.md # Original workshop task description
│
├── .env.example                    # Environment variable template
├── requirements.txt                # Python dependencies
└── README.md

⚙️ Task 1 — Personal Assistant Agent

A versatile, single-agent AI assistant that intelligently selects and invokes custom Python tools based on the user's request.

Architecture

            User
              │
              ▼
    ┌─────────────────┐
    │ Personal        │
    │ Assistant Agent │
    └────────┬────────┘
             │
   ┌─────────┴──────────┐
   ▼         ▼           ▼
 Google   Calculator  Text Analyzer
 Search   (Custom)     (Custom)
 (built-in)

Tools

Tool	Type	Purpose	Input	Output
`google_search`	Built-in	Fetches live information from the web	Search query string	Relevant search results
`calculator`	Custom Python	Evaluates mathematical expressions	Expression string e.g. `"12500 * 18 / 100"`	Computed numeric result
`text_analyzer`	Custom Python	Analyzes text metrics	Any text string	Character count, Word count, Sentence count

Example Interactions

User   → "Calculate 18% GST on ₹25,000."
Agent  → [calls calculator("25000 * 18 / 100")] → "The GST amount is ₹4,500."

User   → "Analyze this paragraph and count its words."
Agent  → [calls text_analyzer("...")] → "Characters: 45 | Words: 8 | Sentences: 1"

Task 1 Agent successfully utilizing custom tools to parse user requests.

🧠 Task 2 — SmartStudy Planner (Multi-Agent Workflow)

A sophisticated multi-agent orchestration system that generates a personalized 10-day exam preparation plan for any subject. It demonstrates all three core workflow patterns taught in the workshop.

Agent Roster

#	Agent	Role
1	Coordinator Agent	Parses user input — extracts subject, topics, level, days, and daily study time
2	Planning Agent	Prepares structured requirements and kicks off parallel analysis
3	Topic Agent	Assigns Priority (High/Med/Low), Difficulty, and Focus for each topic
4	Practice Agent	Recommends problem count and practice type per topic
5	Schedule Agent	Converts available hours into a realistic day-by-day calendar
6	Study Plan Writer	Combines parallel outputs into one cohesive draft study plan
7	Review Agent	Validates the draft and triggers refinement if needed (max 2 loops)

Workflow Architecture

User Input
    │
    ▼
┌───────────────────┐
│  Coordinator      │  ← Parses: Subject, Topics, Level, Days, Hours/day
└────────┬──────────┘
         │ Sequential
         ▼
┌───────────────────┐
│  Planning Agent   │  ← Structures requirements for parallel analysis
└──┬──────┬──────┬──┘
   │      │      │  Parallel
   ▼      ▼      ▼
[Topic] [Practice] [Schedule]    ← All 3 run independently & simultaneously
   │      │      │
   └──────┴──────┘
         │ Merge
         ▼
┌───────────────────┐
│ Study Plan Writer │  ← Aggregates all 3 outputs into one draft
└────────┬──────────┘
         │
         ▼
┌───────────────────┐
│  Review Agent     │  ← Validates: coverage, time, practice, revision, mock test
└────────┬──────────┘
         │
    ┌────┴────────────────────┐
    │ "Needs Improvement"?    │   Loop (max 2 iterations)
    └────────────────────────►─► back to Writer
         │ "APPROVED"
         ▼
    Final 10-Day Study Plan ✅

Workflow Patterns Demonstrated

Pattern	Implementation
Sequential	`Coordinator → Planning Agent → Writer`
Parallel	`Planning → (Topic \| Practice \| Schedule)` simultaneously
Loop	`Writer → Reviewer → Writer` (maximum 2 iterations)

Test Input

I have a Data Structures exam in 10 days.
My current level is Intermediate.
Topics:
- Arrays
- Linked Lists
- Trees
- Graphs
- Dynamic Programming

I can study for 3 hours per day.
Create a preparation plan for me.

Task 2 Multi-Agent Workflow successfully generating a concise 10-day study plan.

🚀 Installation & Setup

Prerequisites

Python 3.10 or higher
Git
A free Groq API key (no credit card required)

1. Clone the Repository

git clone https://github.com/sumitrawat2417/Google-ADK-AI-Agents.git
cd Google-ADK-AI-Agents

2. Create and Activate a Virtual Environment

Windows:

python -m venv venv
.\venv\Scripts\activate

macOS / Linux:

python3 -m venv venv
source venv/bin/activate

3. Install Dependencies

pip install google-adk
pip install "litellm>=1.75.5"

4. Configure Environment Variables

Create a .env file in the root directory:

GROQ_API_KEY=your_groq_api_key_here

Get your free API key at console.groq.com — sign in with Google, no billing required.

💻 Running the Agents

Run both agents using the Google ADK CLI from the project root:

Task 1 — Personal Assistant

adk run Task1_Personal_Assistant

Try these prompts:

Calculate 12500 * 18 / 100 and then tell me how many words are in this sentence: Google ADK makes building AI agents incredibly easy.

Task 2 — SmartStudy Planner

adk run Task2_SmartStudy_Planner

Use the full test input:

I have a Data Structures exam in 10 days. My current level is Intermediate.
Topics: Arrays, Linked Lists, Trees, Graphs, Dynamic Programming
I can study for 3 hours per day. Create a preparation plan for me.

🛠️ Tech Stack

Technology	Role
Google ADK	Agent orchestration framework
LiteLLM	Universal LLM routing layer
Groq API	Free, ultra-fast LLM inference
`qwen/qwen3.8-27b`	Underlying language model (via Groq)
Python 3.10+	Core runtime

🔒 Security Notice

Never commit your .env file or API keys to version control.
The .gitignore is configured to exclude .env by default.
Before sharing or submitting this project, verify that no API keys are present in any file.

📄 License

This project was created as an exploration of the Google ADK (Agent Development Kit). All custom code is open-source and available for educational use.

_{Built with ❤️ by Sumit Rawat · Google ADK Workshop 2026}