FitShot: Multi-Agent AI Personal Fitness & Nutrition Coach

This repository contains a multi-agent AI system built using Google’s Agent Development Kit (ADK).
FitShot acts as a personalized fitness coach that:

• Assesses a user’s fitness level
• Generates a structured weekly workout plan
• Creates a nutrition plan aligned with the workout & goals

All powered through a Sequential Multi-Agent Pipeline.

---

📌 Project Overview

• Built a three-stage reasoning pipeline using ADK SequentialAgent
• Specialized sub-agents handle: Assessment → Workout Planning → Nutrition Planning
• System tailors routines to (beginner/intermediate/advanced)
• Deterministic, structured, explainable behavior
• Fully runs inside a Kaggle Notebook
• Demonstrates state passing, sequential workflows, and multi-agent orchestration

---

🚀 Features

✔ Fitness Assessment

Classifies user into:

• Level: beginner / intermediate / advanced
• Goal: fat_loss / muscle_gain / general

✔ Workout Plan Generator

Produces:

• 7-day weekly routine
• Exercise types (HIIT, strength, mobility, cardio)
• Intensity scaling
• Rest days

✔ Nutrition Plan Generator

Generates:

• Daily calorie target
• Macro breakdown
• Meal ideas
• Nutrition tailored to workout demands

✔ Multi-Agent ADK Pipeline

Uses:

• AssessmentAgent
• WorkoutPlannerAgent
• NutritionPlannerAgent
• All orchestrated by SequentialAgent (FitShotPipeline)

Every sub-agent writes into the shared state via output_key.

---

📐 Architecture

User Message
    ↓
AssessmentAgent  
    ↓
WorkoutPlannerAgent  
    ↓
NutritionPlannerAgent  
    ↓
Final Combined Output

State Passing:

• AssessmentAgent → fitness_profile
• WorkoutPlannerAgent → workout_plan
• NutritionPlannerAgent → nutrition_plan

Agents read state using placeholders:

{fitness_profile}
{workout_plan}

---

📂 Repository Structure

├── fitshot_agent.ipynb        # Kaggle Notebook with full implementation
├── fitshot_agent.py           # Agent definitions (optional)
├── README.md                  # Documentation
└── assets/                    # Architecture diagrams (optional)

---

🎮 Example Run

User Input

“I’m a beginner and want fat loss workouts, preferably HIIT.”

AssessmentAgent Output

{
  "level": "beginner",
  "goal": "fat_loss"
}

WorkoutPlannerAgent Output

{
  "monday": "20-min low-intensity HIIT",
  "tuesday": "Full-body strength",
  "wednesday": "Rest/Mobility",
  "thursday": "HIIT",
  "friday": "Core + Strength",
  "saturday": "Cardio walk",
  "sunday": "Rest"
}

NutritionPlannerAgent Output

{
  "daily_calories": 1800,
  "macros": {"protein": 40, "carbs": 30, "fat": 30},
  "meals": [
    "Oats + berries",
    "Chicken bowl",
    "Salmon + veggies",
    "Green smoothie"
  ]
}

---

🧠 Key Learnings

• Built a real-world concierge-style agent
• Demonstrated ADK SequentialAgent workflow
• Showed clean state passing between agents
• Created an explainable, reproducible multi-agent pipeline
• All runnable inside Kaggle with no backend

---

🛠️ Setup & Usage

1. Clone the repo

git clone https://github.com/<your-username>/fitshot-agent.git
cd fitshot-agent

2. Install dependencies

pip install google-adk

3. Add Google API Key

export GOOGLE_API_KEY="your_api_key"

4. Run the FitShot Pipeline

from google.adk.runners import InMemoryRunner
from fitshot_agent import root_agent

runner = InMemoryRunner(agent=root_agent)
response = await runner.run("I am a beginner wanting fat loss")
print(response)

---

🔮 Future Enhancements

1. Personalized Daily Fitness Routine Delivery

   • Store each user’s long-term fitness profile and history.
   • Instead of generating only weekly plans, deliver daily workout plans automatically.
   • Send nutrition plans + hydration reminders directly to WhatsApp.
   • Deliver a daily motivation quote, aligned with the user’s fitness level and next-day workout intensity.

2. Weekly Progress & Analytics Dashboard

   • Automatically estimate calories burned per workout and compute total weekly burn.
   • Provide a progress forecast based on user consistency (expected time to reach fat-loss / muscle-gain goals).
   • Compare planned vs. completed workouts and adjust next week’s routine dynamically.

3. Smart Fitness Shopping Planner

   • Recommend essential equipment (resistance bands, dumbbells, mats, shoes) tailored to:
     • User level
     • Workout type
     • Budget
   • Maintain a shopping wishlist and upgrade suggestions over time.

4. Healthy Food Discovery & Ordering System

   • Suggest healthy meal options from nearby stores or restaurants.
   • Integrate location-based recommendations:
     • High-protein meals
     • Low-calorie options
     • Smoothies, salads, meal-prep services
   • If the user confirms, the agent can:
     • Prepare the order
     • Interface with delivery services’ APIs (future integration)
     • Track recurring orders (meal-prep weekly routine)

---

📚 References

• Google ADK Documentation
• Gemini API
• ACSM Fitness Guidelines
• Multi-agent workflow patterns (Sequential, Loop, Parallel)

---

✨ Author

Meghana Nanuvala
Master's Student, Computer Science — Indiana University
Interests: AI/ML, Agentic Systems, Fitness Tech, Generative AI