🍽️ Plate Pal

Where AI meets your Appetite

An AI-powered recipe recommendation system built with Streamlit that matches personalised nutritional goals to recipes using cosine similarity — with smart health-aware filtering for medical conditions.

---

✨ Features

• 🔍 Personalised Recommendations — cosine similarity matching against your daily nutritional targets
• 🩺 Health-Aware Filtering — auto-filters recipes for high blood sugar (≥126 mg/dL) or high blood pressure (≥130 mmHg)
• 📂 CSV Upload — works with any compatible recipe dataset
• 🏷️ Nutritional Badges — displays calories, fat, carbs, protein, and match % per recipe
• ⬇️ Export Results — download top recommendations as a CSV
• ⚡ Cached Loading — dataset is cached for fast repeated interactions
• 👤 Session-based Sign In — persistent user profile across interactions

---

🛠️ Tech Stack

Component	Technology
Framework	Streamlit
Data Processing	Pandas
ML / Similarity	scikit-learn
Language	Python 3.8+

---

📁 Project Structure

plate_pal/
├── plate_pal_app.py        # Main Streamlit application
├── requirements.txt        # Python dependencies
├── README.md               # Project documentation
└── Recipe Dataset.csv      # Your recipe dataset (user-supplied)

---

⚙️ Installation

1. Clone the Repository

git clone https://github.com/your-username/plate-pal.git
cd plate-pal

2. Install Dependencies

pip install -r requirements.txt

3. Run the App

streamlit run plate_pal_app.py

---

📦 Requirements

requirements.txt:

streamlit>=1.28.0
pandas>=1.5.0
scikit-learn>=1.2.0

---

📂 Dataset Format

Upload a CSV file with the following required columns:

Column Name	Description
Name	Recipe name
Calories	Caloric content (kcal)
FatContent	Fat in grams
CarbohydrateContent	Carbohydrates in grams
ProteinContent	Protein in grams
RecipeInstructions	Step-by-step cooking instructions

---

🚀 Usage

1. Sign In — enter your name in the sidebar and click Sign In
2. Health Profile — set your age, blood sugar level, and blood pressure
3. Nutritional Goals — input your daily targets for calories, fat, carbs, and protein
4. Upload Dataset — upload your Recipe Dataset CSV
5. Get Recommendations — click the button to view your top personalised recipes
6. Export — optionally download results as a CSV file

---

🧠 How It Works

Cosine Similarity

Both the user's nutritional targets and the dataset values are normalised using the dataset's own min/max range (same scale), then cosine similarity is computed between the user vector and every recipe vector. The top-N closest recipes are returned.

user_norm[i] = (user_value[i] - dataset_min[i]) / (dataset_max[i] - dataset_min[i])

⚠️ This is the critical fix over naive implementations — normalising the user against the dataset scale (not against themselves) ensures meaningful similarity scores.

Health Filtering

Condition	Filter Applied
Sugar ≥ 126 mg/dL	Removes recipes above 60th percentile for carbs & calories
Blood Pressure ≥ 130 mmHg	Removes recipes above 60th percentile for fat content

---

⚠️ Known Limitations

• Blood sugar and BP thresholds are heuristic — not a substitute for medical advice
• Sodium is not used (not in base dataset); fat is used as a proxy for BP filtering
• Best results with large, diverse datasets (1,000+ recipes)

---

📄 License

This project is licensed under the MIT License. Feel free to use, modify, and distribute with attribution.

---

Built with ❤️ using Streamlit & scikit-learn