AI Food Advisor: Your Personal AI Nutritionist 🥗

Navigating dietary choices with a health condition can be confusing and stressful, often leading to reliance on generic, impersonal advice. The AI Food Advisor tackles this challenge head-on, providing an instant, personalized, and data-driven conversational chatbot specifically designed to answer the critical question: "Can I eat this?"

This application moves beyond simple lookups with a powerful "Two-Brain" AI architecture.

1. Conversational "Receptionist" (NLP): Uses fuzzy string matching to effortlessly understand natural language, typos, and phrasing variations.
2. "Expert" Biomarker Engine: A highly scalable, abstract layer that evaluates food safety based on underlying nutritional biomarkers rather than hardcoded rules, making it inherently future-proof for new diseases.

---

🚀 The Architecture (Flowchart)

graph TD
    A[User Input: 'Can I eat rice if I have diabetes?'] --> B(NLP Receptionist)
    
    subgraph NLP Layer [NLP & Entity Extraction]
        B -->|fuzz.WRatio| C{Identify Food}
        B -->|fuzz.partial_ratio| D{Identify Conditions}
        C -->|Extracts| E[Spanish Rice]
        D -->|Extracts| F[Diabetes]
    end

    E --> G(Biomarker Engine)
    F --> G
    
    subgraph Engine Layer [Biomarker Intelligence]
        G -->|Maps 11 Nutrients| H[Compute 6 Biomarkers]
        H -->|Glycemic Impact, Caloric Density, etc.| I{Apply Disease Profile}
        I -->|Compare against Thresholds| J[Risk Score Calculation]
    end
    
    J --> K{Decision}
    K -->|Risk <= Threshold| L[✅ GOOD TO EAT]
    K -->|Risk > Threshold| M[❌ AVOID]
    
    L --> N[Response Generator]
    M --> N
    N --> O[UI: Color-Coded Markdown Response]

Loading

---

🔮 Why This System is "Future-Proof"

Traditional machine learning models (like our fallback XGBoost implementation) require you to retrain an entire model every time you want to support a new disease. If a new dataset arrives, or a new disease needs to be supported, you are forced to re-run your ML pipelines.

The Biomarker Engine changes everything: Instead of mapping Food -> Disease directly, it maps Food -> Biomarkers -> Disease.

1. Universal Biomarkers: It computes 6 universal scores (e.g., Glycemic Impact, Cardiovascular Strain, Inflammatory Index) from 11 raw nutritional features (Calories, Sodium, Fats, etc.).
2. Plug-and-Play Diseases: To add a new disease, you simply define its "Risk Profile" (e.g., Hypertension is 75% Cardiovascular Strain + 25% Inflammatory Index). Zero model retraining is required.
3. Auto-Calibration: When the system starts, it scans the dataset and automatically calibrates the risk thresholds for all diseases.

This abstraction allows the system to instantly support Obesity, Kidney Disease, and PCOD/PCOS without needing a single new ML model.

---

✨ Features

• Conversational Interface: Chat naturally. Say "hello", ask "i love rice but i have diabeties", and the bot understands the context.
• Intelligent Typo Handling: Powered by thefuzz, it easily corrects typos (e.g., "diabeties" -> "diabetes") and finds partial food matches.
• Rich Markdown Explanations: Doesn't just say "Yes" or "No". It explains why based on Biomarker levels (e.g., High Glycemic Impact) and lists key nutrients.
• Beautiful UI: A stunning, modern, glassmorphism UI built with Tailwind CSS, featuring floating animations, typing indicators, and color-coded verdicts.
• Supports 7 Health Conditions: Diabetes, Hypertension, Hyperlipidemia, Thyroid Disorder, Obesity, Kidney Disease, and PCOD.

---

🛠️ Technologies & Libraries

• Backend: Flask (Python)
• NLP / Fuzzy Matching: thefuzz (formerly FuzzyWuzzy), python-Levenshtein
• Data Processing: pandas, numpy
• Machine Learning (Fallback/Validation): xgboost, scikit-learn, joblib
• Frontend: HTML5, Vanilla JavaScript, Tailwind CSS (via CDN)
• Markdown Rendering: Showdown.js

---

⚙️ How to Run Locally

1. Install Dependencies

Ensure you have Python 3.8+ installed. Install the required libraries using pip:

pip install -r requirements.txt

(Note: requirements.txt should include flask, flask-cors, pandas, numpy, thefuzz, python-Levenshtein, xgboost, scikit-learn, joblib)

2. (Optional) Run the Training & Validation Script

If you want to train the fallback XGBoost models and see a side-by-side accuracy comparison between XGBoost and the Biomarker Engine:

python train_nutrition_model.py

3. Start the Backend Server

Run the Flask API:

python main.py

The server will start on http://127.0.0.1:5000.

4. Open the Web App

Simply double-click the index.html file to open it in your web browser. Start chatting with your AI Nutritionist!