🍽️ MealMind Google ADK

Complete 3-Agent Meal Planning System using Google ADK Framework

A production-ready multi-agent meal planning system using Google's Agent Development Kit (ADK) with Sequential Workflow pattern.

---

🎯 Overview

MealMind is an intelligent multi-agent meal planning assistant that transforms the overwhelming challenge of family meal planning into a manageable, safe, and personalized experience.

Every week, millions of families face the same struggle: planning meals that satisfy everyone's dietary needs when one child has severe nut allergies, a parent manages diabetes, and another family member follows a vegetarian diet. Traditional meal planning apps serve single users, but real families need solutions that respect overlapping and conflicting dietary requirements simultaneously.

MealMind solves this using Google's Agent Development Kit (ADK) with a 3-agent sequential workflow powered by Gemini AI:

🎯 Key Capabilities

• Generate 7-day meal plans for entire households (demo shows 3 days)
• Respect health conditions - diabetes, PCOS, high blood pressure, and more
• Handle allergies safely - critical allergen validation prevents dangerous ingredients
• Honor preferences - vegetarian, vegan, dietary restrictions, dislikes
• Optimize cooking time - batch prep, ingredient reuse, advance preparation
• Create grocery lists - automated shopping lists with cost estimates
• Learn over time - memory bank tracks family preferences and favorites
• Budget-conscious - stays within household cost constraints

💡 Real-World Impact

MealMind helps:

• Busy parents reduce decision fatigue and save hours each week
• Caregivers manage complex medical diets safely
• Health-conscious families maintain dietary goals without stress
• Multi-generational households satisfy diverse tastes and needs

🤖 Technical Innovation

Built using Google ADK patterns:

• Multi-agent collaboration - 3 specialized agents working in sequence
• Tool integration - structured functions for household data management
• Memory systems - long-term preference storage and session tracking
• Debug traces - workflow execution visibility via run_debug()
• Safety-critical validation - allergen detection and health compliance

Capstone learning demonstrated:

• Sequential workflow orchestration with Google ADK
• LLM-driven reasoning with Gemini models
• Retry logic and error handling for production readiness
• Memory persistence for continuous learning

---

🏗️ Architecture

┌──────────────────────────────────────────────────────────────────┐
│              GOOGLE ADK SEQUENTIAL WORKFLOW                       │
└──────────────────────────────────────────────────────────────────┘
         │
         ▼
┌────────────────────────┐
│  1. Recipe Generator    │  → Generates 9 recipes with Gemini
└────────┬───────────────┘
         │
         ▼
┌────────────────────────┐
│  2. Nutrition Validator │  → Validates safety & nutrition
└────────┬───────────────┘
         │
         ▼
┌────────────────────────┐
│  3. Schedule Optimizer  │  → Optimizes cooking schedule
└────────────────────────┘

---

📂 Project Structure

MealMindGoogleADK/
├── agents/                              # 3 Specialized Agents
│   ├── __init__.py
│   ├── recipe_generator_adk.py          # Agent 1: Recipe generation
│   ├── nutrition_validator_adk.py       # Agent 2: Nutrition validation
│   └── schedule_optimizer_adk.py        # Agent 3: Schedule optimization
│
├── tools/                               # Custom Tool Functions
│   ├── __init__.py
│   ├── nutrition_lookup.py              # Nutrition data & calculation
│   ├── profile_store.py                 # Household profile management
│   ├── cost_estimator.py                # Cost estimation
│   ├── health_guidelines.py             # Health condition rules
│   ├── schedule_tools.py                # Cooking time analysis
│   └── grocery_tools.py                 # Shopping list aggregation
│
├── memory/                              # Memory & Session Management
│   ├── __init__.py
│   ├── memory_bank.py                   # Per-member preference storage
│   └── session_manager.py               # Session history tracking
│
├── utils/                               # Utility Functions
│   ├── __init__.py
│   ├── display_utils.py                 # Output formatting
│   └── meal_planning_utils.py           # Helper functions
│
├── MEALMIND-FINAL-DEMO.ipynb            # Complete demo notebook with docs
├── orchestrator.py                      # Sequential workflow coordinator
├── demo_complete.py                     # Demo script
├── requirements.txt                     # Dependencies
├── .gitignore
└── README.md                            # This file

---

🚀 Quick Start

Installation

git clone https://github.com/UpasanaGhosh/MealMindGoogleADK.git
cd MealMindGoogleADK

python -m venv venv
source venv/bin/activate

pip install -r requirements.txt

echo "GOOGLE_API_KEY=your_key" > .env

Run Demo

python demo_complete.py

Run Jupyter Notebook

jupyter notebook MEALMIND-FINAL-DEMO.ipynb

---

🔧 3-Agent System

Agent 1: Recipe Generator

File: agents/recipe_generator_adk.py

Purpose: Generates meal recipes using Gemini AI

Capabilities:

• Creates 9 recipes (3 days × 3 meals)
• Considers all household dietary constraints
• Ensures vegetarian compatibility
• Manages allergen avoidance (nuts, dairy, etc.)
• Respects health conditions (diabetes, PCOS)
• Stays within time and budget limits

Model: Gemini 2.5 Flash Lite

Input: Household constraints from profile

Output: JSON array of meal recipes

Agent 2: Nutrition Validator

File: agents/nutrition_validator_adk.py

Purpose: Validates recipes for safety and nutritional adequacy

Checks:

• Allergen Detection - Identifies nuts, dairy, gluten, etc.
• Nutritional Compliance - Validates against health guidelines
• Health Conditions - Ensures diabetes-friendly, PCOS-appropriate meals
• Safety Critical - Flags any violations for critical allergies

Model: Gemini 2.5 Flash Lite

Input: Generated recipes from Recipe Generator

Output: Validated recipes with safety notes

Agent 3: Schedule Optimizer

File: agents/schedule_optimizer_adk.py

Purpose: Optimizes meal preparation schedule

Optimization:

• Time Management - Ensures total time fits household constraints
• Batch Cooking - Identifies prep tasks that can be done together
• Ingredient Reuse - Groups recipes sharing ingredients
• Advance Prep - Suggests tasks for previous day
• Final Formatting - Structures output as actionable JSON

Model: Gemini 2.5 Flash Lite

Input: Validated recipes from Nutrition Validator

Output: Optimized meal plan with schedule

---

💡 Usage

Python Script

import asyncio
from orchestrator import create_orchestrator
from tools import create_household_profile, add_family_member

async def generate_plan():
    # Setup household
    create_household_profile("family_01", "My Family", 45, 150.0)
    add_family_member("family_01", "Parent", 35, "vegetarian", "nuts", "diabetes")
    
    # Generate meal plan
    orchestrator = create_orchestrator(api_key="your_key")
    result = await orchestrator.generate_meal_plan("family_01", days=3)
    
    print(result)

asyncio.run(generate_plan())

Jupyter Notebook

The complete workflow is demonstrated in MEALMIND-FINAL-DEMO.ipynb with:

• Comprehensive inline documentation
• Step-by-step explanations
• Interactive execution
• Visual output formatting

---

📓 Kaggle Notebook

File: MEALMIND-FINAL-DEMO.ipynb

23 cells demonstrating complete workflow:

• 📦 Dependency installation and imports
• 🔑 API configuration and memory setup
• 🛠️ Tool and data structure definitions
• 👥 Household profile creation
• 🤖 Agent creation and configuration
• 🔄 Sequential workflow assembly
• ▶️ Meal plan generation
• 📊 Output parsing and organization
• 📋 Display with detailed analysis
• 💾 Memory bank operations
• ✅ Comprehensive summary

Ready for Kaggle!

---

🎓 Capstone Features

✅ Multi-Agent System

• 3 specialized agents with distinct responsibilities
• Sequential coordination pattern
• Google ADK framework implementation
• Clean separation of concerns

✅ Real-World Application

• Family meal planning with multiple constraints
• Critical allergen safety (nuts, dairy)
• Health condition management (diabetes, PCOS)
• Budget and time optimization
• Preference learning via memory bank

✅ Production Ready

• Retry configuration for API failures
• Error handling and validation
• Modular architecture
• Comprehensive documentation
• Session management
• Memory persistence

---

🔑 Google ADK Patterns

Sequential Workflow:

workflow = SequentialAgent(
    name="meal_planning",
    sub_agents=[recipe_agent, nutrition_agent, schedule_optimizer_agent]
)

InMemoryRunner:

runner = InMemoryRunner(agent=workflow)
result = await runner.run_debug(prompt)

Retry Config:

retry_config = types.RetryOptions(
    max_attempts=5,
    backoff_base=7,
    initial_delay=1
)

Agent Creation:

agent = LlmAgent(
    name="recipe_generator",
    model=Gemini(model="gemini-2.5-flash-lite", api_key=API_KEY),
    instruction="Generate recipes...",
    tools=[]
)

---

🔍 Key Features

Constraint Satisfaction

Successfully handles multiple competing constraints:

• Vegetarian-friendly
• Nut-free (critical allergy)
• Low-GI for PCOS
• Low-carb/sugar for diabetes
• 45-minute time limit
• $150 budget constraint

Agent Specialization

Each agent focuses on one responsibility:

• Recipe Generator: Creativity and meal variety
• Nutrition Validator: Safety and compliance
• Schedule Optimizer: Efficiency and timing

Memory & Learning

• Per-member preference tracking
• Session history for context
• Favorite meal storage
• Dietary pattern analysis

---

📈 Demo Household

Demo Configuration:

Household: "Demo Family"
Budget: $150
Time: 45 minutes

Members:
  - Alice (8): Vegetarian, Allergic to nuts
  - Bob (35): Diabetic, Low-carb
  - Carol (32): PCOS, Low-GI

This demonstrates real-world complexity where multiple dietary needs must be satisfied simultaneously.

---

🛠️ Development

Project Dependencies

Key packages:

• google-genai>=1.19.0 - Google ADK framework
• python-dotenv - Environment management
• jupyter - Notebook support

Adding New Agents

1. Create agent file in agents/ directory
2. Define using LlmAgent class
3. Add to workflow sequence in orchestrator.py
4. Update documentation

Adding New Tools

1. Create tool function in tools/ directory
2. Define input/output schema
3. Register with appropriate agent
4. Add documentation and tests

---

📝 License

MIT License

---

🎯 Future Enhancements

• Observability & Monitoring - Structured logging, metrics dashboards, trace analysis, and performance monitoring for production deployment
• Persistent database for meal history
• User authentication system
• Web UI for easier interaction
• Integration with grocery delivery APIs
• Expanded recipe database
• Meal rating and feedback system
• Mobile app development
• Advanced cost estimation
• Regional cuisine support
• Seasonal ingredient optimization

---

Built with Google ADK • 3-Agent System • Production-Ready • Capstone Project