Dosha Prediction AI 🌿

AI-powered Ayurvedic Dosha prediction app using XGBoost + FastAPI + React. Discover your unique constitution and receive personalized wellness guidance.

---

🔗 Links

Resource	Link
🚀 Live App	doshapredictor.vercel.app
📓 Kaggle Notebook	classifier-model-for-dosha-assessment
💻 GitHub Repository	csxark/Dosha-Predictor

---

Project Structure

dosha-prediction-ai/
├── backend/
│   ├── main.py           # FastAPI app + prediction engine
│   ├── model.pkl         # Your trained XGBoost model ← PUT IT HERE
│   ├── requirements.txt
│   ├── Procfile          # Railway deployment
│   └── .env.example
└── frontend/
    ├── src/
    │   ├── components/
    │   │   ├── Landing.jsx
    │   │   ├── Assessment.jsx
    │   │   ├── Predicting.jsx
    │   │   ├── Results.jsx
    │   │   └── Analytics.jsx
    │   ├── data/questions.js
    │   ├── lib/appwrite.js
    │   ├── App.jsx
    │   └── index.css
    ├── .env.example
    └── vercel.json

---

Step 1 — Backend Setup

Place your model

Copy your model.pkl into backend/:

cp /path/to/model.pkl backend/model.pkl

Local development

cd backend
pip install -r requirements.txt
uvicorn main:app --reload --port 8000

Test it:

curl -X POST http://localhost:8000/predict \
  -H "Content-Type: application/json" \
  -d '{
    "bodyFrame": "Thin and Lean",
    "paceOfWork": "Fast",
    "bodyEnergy": "Low",
    "hungerPattern": "Skips Meals",
    "hairType": "Dry",
    "sleepPattern": "Light Sleeper",
    "mentalActivity": "Restless",
    "voiceQuality": "Fast",
    "jointStructure": "Light",
    "skinType": "Dry",
    "bodyOdor": "Mild"
  }'

Deploy to Railway

1. Go to https://railway.app → New Project → Deploy from GitHub
2. Set root directory to backend/
3. Add environment variable: MODEL_PATH=model.pkl
4. Railway will auto-detect Procfile and deploy
5. Note your Railway URL (e.g. https://dosha-backend.up.railway.app)

Important: Your model.pkl must be committed to the repo, or uploaded via Railway's volume storage.

---

Step 2 — Appwrite Setup (Database)

1. Go to https://cloud.appwrite.io → Create Project
2. Create a Database → note the Database ID
3. Create a Collection called assessments → note Collection ID
4. Add these attributes to the collection:
   • assessmentId → String (255)
   • timestamp → String (64)
   • predictedDosha → String (32)
   • confidence → Float
   • secondaryDosha → String (32)
   • answers → String (4096)
5. Set Collection permissions: Any can Create and Read (for anonymous storage)

---

Step 3 — Frontend Setup

Local development

cp .env.example .env.local
# Fill in your values in .env.local
npm install
npm run dev

Environment variables

VITE_API_URL=https://your-railway-backend.up.railway.app
VITE_APPWRITE_ENDPOINT=https://cloud.appwrite.io/v1
VITE_APPWRITE_PROJECT_ID=your_project_id
VITE_APPWRITE_DATABASE_ID=your_database_id
VITE_APPWRITE_COLLECTION_ID=your_collection_id

Deploy to Vercel

1. Push frontend folder to GitHub
2. Go to https://vercel.com → New Project → Import repo
3. Set root directory to frontend/
4. Add all 5 environment variables in Vercel dashboard
5. Deploy!

---

Model Integration Notes

The backend uses label encoding (ordinal) for all features. The encoding maps in main.py must match your training preprocessing exactly.

If your model used one-hot encoding instead, update encode_input() in main.py:

def encode_input(data: AssessmentInput) -> pd.DataFrame:
    row = {field: getattr(data, field) for field in FEATURE_ORDER}
    df = pd.DataFrame([row])
    df_encoded = pd.get_dummies(df)
    # Align with training columns
    training_cols = model.get_booster().feature_names
    df_encoded = df_encoded.reindex(columns=training_cols, fill_value=0)
    return df_encoded

Tech Stack

Layer	Technology
Frontend	React + Vite + Tailwind CSS v4
Animations	Framer Motion
Charts	Recharts
Backend	FastAPI + Uvicorn
ML Model	XGBoost via Joblib
Database	Appwrite Cloud
Deployment	Vercel (frontend) + Railway (backend)

---

Notes

• No authentication — fully anonymous
• No PII collected — only answers + prediction stored
• Fallback mode — if model.pkl fails to load, a rule-based algorithm runs
• Fallback UI — if the API is down, a mock result is shown so the UI never breaks