FitAI — AI-Powered Fitness & Nutrition Coach

Built by Adarsh Kumar Pandey · B.Tech CSE (AI & ML) · JSSATEN, Noida

A production-ready fitness web app combining Machine Learning and Large Language Models to deliver personalized Indian diet and workout plans — all in a clean, dark UI.

🔗 Live Demo:(https://m3g4bwehqnud6ck35fgbwd.streamlit.app/)

Features

Plan Generator

Predicts daily calorie target using a trained Random Forest Regressor
Classifies fitness level (Beginner / Intermediate / Advanced) using Random Forest Classifier
Generates a complete 7-day Indian meal + workout plan using LLaMA 3.3 via Groq API
Displays a real-time macro split bar (Protein / Carbs / Fat %)

BMI & Health Score

Visual BMI bar with category detection (Underweight / Normal / Overweight / Obese)
Composite health score (0–100) based on BMI, activity level, and age
Personalised diet + lifestyle recommendations

Indian Food Lookup

Database of 24+ common Indian foods with full macro breakdown
Search by name, add to daily log, track total calories + macros in real time

Progress Tracker

Every plan generation is auto-saved locally
Session history with goal, weight, and macro breakdown
SVG calorie trend graph across sessions

Tech Stack

LayerTechnologyML Modelsscikit-learn (Random Forest Regressor + Classifier)LLMLLaMA 3.3 70B via Groq APIFrontendStreamlit + Custom CSSDataSynthetic dataset (2000 samples, Harris-Benedict BMR formula)LanguagePython 3.10+

Project Structure

fitai/
├── app.py                  # Main Streamlit application
├── generate_data.py        # Synthetic dataset generation
├── train_model.py          # ML model training
├── requirements.txt        # Dependencies
├── .streamlit/
│   └── secrets.toml        # API key (local only, never pushed)
├── calorie_model.pkl       # Trained regression model
├── fitness_model.pkl       # Trained classifier model
├── le_fit.pkl              # Label encoder
└── le_goal.pkl             # Label encoder

Local Setup

# 1. Clone the repo
git clone https://github.com/YOUR_USERNAME/fitai.git
cd fitai

# 2. Create virtual environment
python -m venv venv
venv\Scripts\activate        # Windows
# source venv/bin/activate   # Mac/Linux

# 3. Install dependencies
pip install -r requirements.txt

# 4. Generate data and train models
python generate_data.py
python train_model.py

# 5. Add your Groq API key
mkdir .streamlit
echo '[secrets]\nGROQ_API_KEY = "gsk_your_key_here"' > .streamlit/secrets.toml

# 6. Run the app
streamlit run app.py

ML Model Details

ModelTypeAlgorithmMetricCalorie PredictorRegressionRandom Forest (200 trees)MAE ~15 kcalFitness ClassifierClassificationRandom Forest (200 trees)Accuracy ~99%

Dataset: 2,000 synthetic samples generated using the Harris-Benedict BMR formula with TDEE multipliers. Goal-based calorie adjustments: +300 kcal (muscle gain), −400 kcal (fat loss).

Features used: Age, Weight, Height, Gender, Activity Level, Goal

Made by Adarsh Kumar Pandey